How to Improve Brand Visibility in AI Search Engines (2027)

Table of Contents

1. What strategies improve brand visibility in AI search engines?
2. How do AI search engines decide which brands to mention?
3. How is AI search visibility different from regular SEO?
4. Which AI visibility tactics have real evidence behind them?
5. Which popular tactics failed when someone actually tested them?
6. How to improve brand visibility in AI search results, platform by platform
7. Why does ChatGPT recommend your competitor instead of your brand?
8. How to improve brand visibility in AI generated answers when you already rank on Google
9. How do you give AI tools something original to cite?
10. What do you do when an AI tool says something wrong about your brand?
11. Do customer reviews change what AI tools say about your brand?
12. How long does each tactic take to show up in AI responses?
13. What does this cost to run in-house?
14. Which AI visibility metrics should you track: mentions, citations or referrals?
15. How do you prove AI visibility work made money?
16. What do Search Console, Bing Webmaster Tools and GA4 actually show about AI visibility?
17. How do you check your brand’s AI visibility for free?
18. How many prompt runs do you need before a change in AI visibility is real?
19. Who should own AI visibility work inside a company?
20. How to improve AI search visibility for enterprise software
21. How do you make your brand readable to AI shopping and booking agents?
22. Which of these tactics could actually get your brand in trouble?
23. How to ensure company visibility in AI: the first 30 days
24. Improve brand visibility in AI search FAQs

What strategies improve brand visibility in AI search engines?

In short, six things move the needle, sorted by how strong the evidence actually is:

What do the most repeated AI visibility tips get wrong?

And the honest counterpart. By comparison, these are the tactics recommended most often that have the least evidence:

We have set out the reasoning and the sources for every row further down. If you only take one thing from this page, take this: the gap between what people recommend and what anyone has tested is the widest we have seen in any marketing discipline. As a result, a lot of budget goes into that gap.

How do AI search engines decide which brands to mention?

They do not all work the same way, and the differences matter more than the similarities.

To begin with, three of the five major surfaces retrieve live from a search index at the moment you ask. Google AI Overviews and AI Mode retrieve from Google’s index. Similarly, Microsoft Copilot retrieves from the Bing index. Perplexity runs its own crawler and index. In contrast, ChatGPT combines a live search layer with what the model already absorbed in training. Claude searches the live web when the conversation calls for it and cites what it finds.

That split also explains a lot of confusing advice. For example, a tactic that changes what a live retriever finds can work within days. However, a tactic aimed at what the model “knows” is aimed at a training run you cannot schedule.

What each platform states on the record

We read the primary documentation rather than other people’s summaries of it. Here is what each company actually commits to in writing, so you can check it yourself.

Then read the second column again. In other words, every one of the five says, in its own documentation, that it does not publish the mechanism. That is a documented absence, not an oversight, and it is the single most important fact on this page.

What no platform will tell you

Because the mechanism is undisclosed, the industry fills the gap with correlation studies, and the correlation studies disagree with each other in ways that are rarely acknowledged.

On whether ranking on Google predicts being cited by AI, there are credible numbers on both sides. One agency measuring 400 plus keywords across 16 clients found that when a client ranked on Google page one, they appeared in ChatGPT and Perplexity answers for the same keyword 77% of the time, rising to 82% in the top three. In contrast, Ahrefs, looking at the question from the URL side, found only about 12% of AI cited URLs ranked in Google’s top 10. Semrush reported a similar pattern in a June 2025 study: the pages ChatGPT search cites rank in position 21 or lower for related queries almost 90% of the time. Meanwhile, Seer Interactive found 87% of SearchGPT citations matched Bing’s top results. One practitioner reports a page earning roughly 1,200 Microsoft Copilot citations while sitting at Google average position 22.4.

However, those findings are not actually incompatible, and understanding why is useful. The high numbers measure whether a brand appears when it already ranks, which is a brand-level question. In contrast, the low numbers measure which specific URLs AI tools cite, which is a page-level question. Finally, the Copilot outlier is a different retriever reading a different index. A brand can be highly visible in AI answers while almost none of its individual pages are the cited URL.

How is AI search visibility different from regular SEO?

AI search visibility measures whether an AI answer names or cites your brand, while regular SEO measures where your pages rank. The two overlap, because Google says its generative AI features are rooted in its core Search ranking and quality systems. However, the unit of success, the reports and the stability of results all differ, so you need both scorecards.

The audience is now large enough to justify the second scorecard. At I/O on May 19, 2026, Google said AI Overviews has over 2.5 billion monthly active users and AI Mode has passed 1 billion. In addition, OpenAI said on August 31, 2026 that its ad-supported free tier helps keep ChatGPT available to more than 1 billion weekly active users.

Ranking still helps, but it is not the whole story. For example, the Semrush study above found that the pages ChatGPT search cites usually rank outside the top 20. So treat AI visibility as a layer on top of SEO rather than a replacement for it. Our guide to what Google documents for AI search covers the Google side in more depth.

Which AI visibility tactics have real evidence behind them?

So we sorted every tactic we encountered into three tiers. In other words, the tier is about evidence quality, not about how useful the tactic feels.

Tier 1: documented by the platform itself

These are things a platform has put in writing. As a result, they are the safest place to spend first.

1. Allow the AI crawlers you want, and verify in your server logs that they are actually getting through. In fact, this is the only item on this page with a one line fix and an invisible failure mode. One practitioner found his own bot protection had been silently blocking AI crawlers by default, and estimates roughly half of major sites block at least one AI crawler without ever deciding to.
2. Stay indexed and snippet eligible in Google. Indeed, Google states this plainly as the precondition for AI Overviews and AI Mode.
3. Claim Bing Webmaster Tools and turn on the AI Performance report. After that, you get four metrics from Microsoft: Total Citations, Average Cited Pages, Grounding Queries and page-level citation counts. Microsoft also states its own limits, which we respect here: grounding queries are a sample, the citation count reflects how often pages are cited rather than importance or ranking, and the data is aggregated across Copilot and Bing AI summaries so you cannot separate them.
4. Set the Preferred Sources flag in Search Console if you are eligible. That matters because Google documents that Preferred Sources can appear in AI Mode and AI Overviews.
5. Do not produce content mainly to manipulate AI responses. In fact, Google documents this as scaled content abuse. So treat it as written policy, not as a caution.

Tier 2: measured in a published test

Unlike Tier 1 items, these have a named study with a method and an effect size behind them. Treat the numbers as directional, since most were measured in a benchmark rather than in live production.

Two caveats before you rely on this study. The main benchmark ran on GPT-3.5 answering from Google’s top five results, with Perplexity used only as a secondary real-world check, so it may not transfer cleanly to today’s engines. And the same body of work found the gain is not uniform: lower-ranked pages saw large gains from adding sources and statistics while already top-ranked pages saw a decline from the same edit. So if you are already the top result, adding citation density may cost you.

Do earned mentions beat owned content?

In addition, this Tier 2 finding compares earned and owned content head to head. Specifically, in a 30-day cold-start test on a link building SaaS with no baseline visibility, across 15 keywords and six AI platforms, the brand went from zero to 298 appearances. Earned placements produced 85.8% of those citations, while owned content produced 14%. The brand’s own listicle, in fact, took 18 days to earn its first citation, the slowest of any source in the test. The author’s own conclusion, revising her earlier position, was that owned listicles are a foundation and not a growth lever.

Tier 3: repeated everywhere, tested by no one

Schema markup, content freshness, author credentials and community posting are among the most repeated recommendations on this topic. However, none of the four has a published test showing it raises AI citations, and the two public schema tests covered below found no lift.

That does not make them worthless. Instead, it makes them unpriced. We also applied the same evidence test to the wider discipline in answer engine optimization. So if someone is charging you for them as a distinct AI visibility service, you are paying a premium for an untested hypothesis.

Which popular tactics failed when someone actually tested them?

Someone has tested each of these in public, so here is what the tests found.

Schema markup

In short: two independent tests, different methods, same answer.

First of all, a seven-platform experiment run between December 2025 and March 2026 did three things. It asked seven AI platforms to fetch the raw schema from a page: only Gemini retrieved the correct JSON-LD, six of seven failed, and Google AI Mode hallucinated a Service schema that did not exist on the page. It then measured brand coverage across 319 US market prompts before and after implementing schema, and the platforms moved in opposite directions: Google AI Overviews up 611%, Google AI Mode up 42%, ChatGPT down 71%, Gemini down 35%. Finally it added FAQ schema containing information available nowhere else online, and no platform answered using it. In the end, the stated conclusion was that schema did not appear to directly influence AI search citation behavior in a measurable way for most platforms.

Similarly, a matched difference in differences test across 1,885 pages reported Google AI Overviews down 4.6%, Google AI Mode up 2.4%, ChatGPT up 2.2%, and concluded that adding schema did not increase AI citations on any platform. A correlation does exist in larger datasets, where cited pages are roughly three times more likely to carry JSON-LD, but that is attributed to the kind of sites that bother with structured data rather than to the markup itself.

However, there is a complication, and we think it is a real one: Microsoft’s Fabrice Canel has publicly confirmed that schema helps Microsoft’s models understand content for Copilot. But neither controlled test isolated Copilot. So the strongest published tests and the strongest published platform statement are talking about different engines. Overall, our read is that schema is cheap and has other justifications. Still, nobody should sell it to you as an AI visibility lever with a number attached.

llms.txt

To begin with, a 90-day server log study on a site with llms.txt implemented recorded 62,100 plus AI bot visits in total. However, only 84 of them touched the llms.txt file, roughly 0.1%. The average page on the site received around 265 AI bot visits, so the file performed about three times worse than an average page. Similarly, Ahrefs, looking across 100,000 domains, reported 97% of llms.txt files got no requests at all. Moreover, Google has stated on the record that it does not support the file and is not planning to.

There is one published counterexample and it deserves to be stated fairly, because almost nobody cites it. A developer submitted his llms.txt through the Search Console URL Inspection tool on February 2, 2026, and documented with screenshots and Cloudflare access logs that the file was crawled the same day, appeared as the number one source in Google AI Mode for a German language query about his company the following day, persisted on 5 February, was hit by six named bots between 14 and 18 February, and was cited across three of four topic queries on 20 February. He is also explicit that his tracked keyword positions did not move at all during the 18-day window. Still, it is one site, one language, and heavily branded queries, and he does not control for that.

There is also a counter-argument about the test itself. One developer argues that measuring crawler logs answers the wrong question, because llms.txt is aimed at agentic tools that a user points at a specific URL, not at discovery crawlers. However, nobody has yet published a test built around that claim.

The cats.txt test, and why you should apply it to every claim on this topic

In August 2026 a technical SEO invented cats.txt, a joke standard listing office cats with job titles, breeds and a “PurrLevel” score, published a spec, and wrote it up as the missing standard for SEO and GEO. Soon, other practitioners adopted the joke, and someone built a site for it. Then PerplexityBot, GPTBot, ClaudeBot and Googlebot all crawled it, and Google indexed it. Google’s AI Overview then reported that a fictional cat named Odd was a “Render Cat” with a PurrLevel of 5 out of 7 who chases the cursor and pounces on stray pixels.

The point, however, is methodological, and it is the most useful idea in this entire topic. In fact, cats.txt satisfied all four proofs that people routinely offer for llms.txt working: bots crawled it, Google indexed it, LLMs repeated invented content from it, and ChatGPT endorsed it as a ranking tactic. Cats.txt is obviously worthless. Therefore those four proofs prove nothing.

So use it as a filter. When someone shows you evidence that an AI tactic works, ask whether cats.txt would have passed the same test. If it would, then the evidence is not evidence.

Listicles and comparison pages

This one, in particular, is a live example of a tactic decaying under you. In August 2026 listicles fell from 15.77% to 7.80% of ChatGPT citations, a 50.5% relative decline, and comparison pages fell from 9.08% to 6.17%. Meanwhile, product pages rose to 16.39% of retrieved pages over the same period. The reported timeline points at a ChatGPT model update on 6 August and a Google spam update rolling out 18 to 21 August.

What makes this instructive, though, is the sequencing. A well-run 14-day study published weeks earlier had recommended comparison and alternatives pages on the strength of real data. Indeed, the data was real. The tactic still collapsed. As one write-up put it, a page that exists only to catch a comparison query has no audience underneath it, so when retrieval behavior shifts nothing holds it up.

Scaled AI content

For example, one experimenter launched two fresh domains in April 2026 with roughly 1,000 AI-generated posts each. Then one collapsed in Google from 1,629 to 15 daily impressions on 9 April; the other from 859 to 85 on 26 June. Search Console showed “No issues detected” under Manual actions the whole time, which is the detail worth keeping: algorithmic suppression does not produce a manual action, so the report you would naturally check tells you nothing.

The same experiment also produced the most interesting AI visibility data point we found anywhere, which we come back to under Copilot below.

How to improve brand visibility in AI search results, platform by platform

AI platforms retrieve from different places, so the same tactic does not move each one the same way. So here is what to do on each.

Google AI Overviews and AI Mode

Google’s own documentation is unusually direct here, so it is worth taking at face value: no special file, no special markup, no AI text file. In short, Google must be able to index the page and show it as a snippet. Normal SEO practice is therefore the lever. Google adds one genuine steer, which is that a unique point of view helps because its systems look across a variety of sources, though it does not quantify that. In addition, our separate guide covers how to optimize content for Google AI Overviews in more depth.

The practical implication is uncomfortable for the AI SEO industry: for the largest AI surface by reach, the recommended work is the work you should already be doing. Our technical SEO checklist covers the indexation and crawl side of that, and AI search optimization covers what Google actually documents in more depth.

ChatGPT

Two separate controls, and mixing them up is a common and expensive mistake. Blocking GPTBot removes your content from OpenAI’s training data. However, it does not remove you from ChatGPT search. Instead, OAI-SearchBot is the one that governs search inclusion, and blocking it removes you from ChatGPT search entirely. Even so, plenty of sites have blocked the wrong one, in both directions.

OpenAI states that it ranks results using multiple factors, but it declines to name them. So the honest ChatGPT playbook is: allow OAI-SearchBot, verify in logs that it is getting 200 responses, and then work on being mentioned by the sources ChatGPT retrieves from, because you cannot work on the ranking layer you cannot see.

One caution on measurement, though: crawl volume is not a proxy for citation value. For example, in the two-site experiment above, ChatGPT-User crawled 4,808 times and cited zero times. If a vendor reports rising AI crawler traffic as a result, that is activity, not outcome. We go deeper on this distinction in citations versus mentions and on measurement specifically in how to track brand mentions in AI search.

Microsoft Copilot

Copilot gets far less attention than ChatGPT or Google in AI visibility advice, and it deserves more of yours.

Three things make it worth your attention.

First of all, Bing Webmaster Tools gives site owners free first-party citation counts and grounding queries for Copilot. Google Search Console now reports AI Overviews and AI Mode impressions by page as well. By comparison, OpenAI offers no public report.

Second, the published first-party numbers suggest Copilot cites far more readily than the surfaces everyone optimizes for. One operator on a DR-25 domain found 671 Copilot citations across 90 days that he had no idea existed, climbing to roughly 1,200 by late May 2026, while his measured citation rate on ChatGPT, Perplexity and Claude was around 30%. Similarly, an agency reported 6,700 Copilot citations in 90 days from 111 unique grounding queries, with a page at Google average position 22.4 earning about 1,200 of them.

Third: in the two-site scaled content experiment, the site that Google had deindexed entirely went on to earn 2,718 Microsoft Copilot citations, 69% of its 3,934 total, while earning exactly zero from ChatGPT and zero from Google AI and Gemini, over the same period, on the same content. Meanwhile, its Bing impressions grew to 91,104 in July against zero Google impressions.

Of course, we are not presenting that as a strategy. Instead, it is an observation that Copilot’s bar for citation is materially different from ChatGPT’s and Google’s, measured on identical content at the same time. So if your category has any Microsoft-heavy audience at all, Copilot is where your effort is currently least contested.

Why do Copilot citations cluster on a few pages?

Two practical notes. In particular, citation concentration on Copilot is extreme in every dataset we found: five pages accounted for 74.6% of citations in one, with the homepage alone at 32.3%; three pages accounted for 87.25% in another. So a Copilot strategy is a small-number-of-pages strategy, not a publish-more strategy. Also, Microsoft is the one platform that has said schema helps its models, which is why we did not write schema off entirely earlier.

Perplexity

Perplexity runs its own crawler and index, so PerplexityBot access is the gate. Our Perplexity SEO guide covers the full setup. The generative engine optimization benchmark that produced the citation, statistic and quotation numbers was tested primarily against Perplexity, so those content edits were also tested here, with smaller and mixed gains.

Set expectations on scale, though. In one six-platform 30-day test, for example, Perplexity produced 4 appearances out of 298. Perplexity is worth tracking, and we have written a full guide to Perplexity rank tracking, but for most B2B and ecommerce categories it is not where the volume is.

Claude

Anthropic documents three separate crawlers, each controlling a different behavior: ClaudeBot for training, Claude-User for fetches triggered by a user’s question, and Claude-SearchBot for search indexing. If you want to appear in Claude’s cited answers while staying out of training data, that is a supported configuration and you should set it deliberately rather than block everything with one rule.

Anthropic also documents that its web search cites sources with a 150 character cited text cap. That is a small but real structural hint: a self-contained claim that survives being quoted in 150 characters is easier to cite than the same claim spread across a paragraph.

Why does ChatGPT recommend your competitor instead of your brand?

This is the most common phrasing of the question on Quora, and it appears there in at least six near-identical variants. Almost no published article answers it with a diagnostic procedure, so here is one. Work through it in order, because each step rules out the ones below it.

Why does the same prompt name different brands each time?

Step 5, in fact, is not a joke. In a study of 2,961 runs by 600 volunteers across ChatGPT, Claude and Google AI, the probability of getting an identical brand list across 100 runs was under 1 in 100, and under 1 in 1,000 for the same list in the same order. In other words, a competitor appearing once and you appearing the next time is the normal behavior of the system, not a diagnosis.

How to improve brand visibility in AI generated answers when you already rank on Google

This is the frustrating case, and the Quora wording for it is blunt: why do some products never show up in ChatGPT or AI search results, even if they rank on Google.

In fact, there are two different invisibility problems, and they need different work.

Retrieval invisibility means the model can find you but does not choose you. You rank, you are indexed, the crawlers get through, and you still are not cited. So the lever here is association: which sources mention you, who you appear alongside, and how recently. The cold-start experiment above found exactly this, and the author’s summary is worth quoting as a principle rather than a statistic: LLM visibility depends more on associations than on page ranking.

Training invisibility, on the other hand, means the model does not know you exist as an entity. As a result, it will not volunteer you in an answer that does not trigger a live search. You cannot fix this on a schedule, because you do not control training runs. What you can do is make the entity unambiguous everywhere it appears, so that when a training run does happen, or when a live retrieval happens, there is one consistent story about who you are and what you do.

A useful diagnostic someone posed and nobody answers in published content: if you Googled your brand today and then asked ChatGPT about it, would the two answers tell the same story? In short, where they diverge is where your entity is unclear.

How do you make your brand entity unambiguous?

You make a brand unambiguous by describing it the same way everywhere a model might read about it. First of all, write one plain sentence that says who you are, what you sell and where. Then reuse it on your site and your profiles, so live retrieval and any future training run find one consistent story.

• One description: use the same sentence on your homepage, About page, social profiles and directory listings.
• One name: spell the brand the same way every time, and link your legal and trading names where both exist.
• Organization structured data: Google says it can help Google understand your organization’s details and disambiguate it in search results, so add it with sameAs links to your official profiles.
• Old descriptions: update outdated bios, directory entries and partner pages, because live retrieval can only find what is published.

Note the limit, though. Google’s AI guide also says structured data is not required for its generative AI features, so treat the markup as housekeeping rather than a citation lever. For the full method, see our guide to entity SEO.

How do you give AI tools something original to cite?

You give AI tools something original to cite by publishing first-party data that other people can check, quote and link to. Google’s AI optimization guide says a first-hand review provides a unique perspective. Also, the GEO paper above found that statistics and citations gave some of the largest gains, and the cold-start test showed earned mentions doing most of the work.

What did we do with our own Google Ads benchmark?

Here is a worked example from Hustle Marketers, described only as the steps we took. In October 2026 we published a report, “Google Ads Performance Benchmarks by Campaign Type”, on Zenodo, an open research repository.

1. Permanent identifier: the record has the DOI 10.5281/zenodo.23719473, so a citation does not depend on one page URL.
2. Named, verifiable author: the record lists Ishant Sharma with his ORCID iD, 0009-0000-3421-5104, and Hustle Marketers as his affiliation.
3. Stated method and sample: the abstract names 22 agency-managed accounts from October 2025 to September 2026. It also says the figures describe one agency’s portfolio rather than the whole market.
4. Reusable data: the anonymised CSV sits next to the PDF under a CC BY 4.0 license.
5. Citing page with matching schema: our PPC statistics page cites the paper in visible text. Its structured data also links the DOI, the ORCID iD and the license.

We have not measured any AI citation result from this yet, so treat it as a method rather than proof. Still, it gives each number a short, self-contained form. For example, “In 22 agency-managed Google Ads accounts, Performance Max took a median 53.1% of ecommerce spend” fits inside the 150-character cited text cap that Anthropic documents for Claude’s web search.

What should a citable research asset include?

Use this checklist before you publish your own research. In each row, the example column shows how our benchmark handled it.

What do you do when an AI tool says something wrong about your brand?

We went looking for a documented case of a brand getting an AI platform to correct a false claim through an official channel. However, we did not find one: not a single verified first-person account. In fact, that absence is itself the finding, and every page that tells you to “address it at the source” without saying how is glossing over it.

So here is what is actually available, in order of how much control you have.

What if an AI tool quotes a page that no longer exists?

One anomaly worth knowing about because it breaks the usual advice: a researcher documented ChatGPT citing a Wikipedia page that had been deleted months earlier and was never indexed by Google. So if a model is drawing on something that no longer exists on the live web, fixing the live web does not reach it.

Our honest position: treat wrong AI claims as a reputation problem with a slow fix, not a technical problem with a fast one. So fix the sources, publish the correction, and then monitor. Meanwhile, anyone promising to get an AI platform to retract something is promising a mechanism nobody has publicly demonstrated.

How do you improve how AI tools describe your brand?

You improve how AI tools describe your brand by changing the sources the answers draw on, not the answers themselves. No platform documents a sentiment score you can influence directly. However, every retrieval surface on this page cites published pages, so the tone of those pages is the lever you control.

• Log the exact sentence: record how each answer describes you, not only whether it names you.
• Trace it: open the cited pages and find where a negative or outdated framing comes from.
• Fix or answer it: correct what you control, ask publishers to update what they control, and respond to real complaints in public.
• Re-test on a schedule: rerun the same prompts monthly, because one answer is not a trend.

Do customer reviews change what AI tools say about your brand?

For shopping answers, yes, and it is on the record. OpenAI’s help page on shopping with ChatGPT search says ChatGPT may consider price, reviews and ease of use when it picks products. It also says review summaries come from reviews on public websites. Outside shopping, though, no platform documents reviews as a factor in brand mentions.

So treat reviews as Tier 1 for products and as Tier 3 for everything else. Here is what each source actually documents.

Either way, the practical move is the same. First of all, ask real customers for reviews on the platforms your buyers already use, and never write or buy them. Then, if you sell products, get those reviews into Merchant Center. For example, our guide to syncing Judge.me reviews to Google Merchant Center shows one way to do it.

How long does each tactic take to show up in AI responses?

In general, published articles give one anecdotal number for the whole discipline, usually somewhere between six weeks and twelve months. That is not useful, because the timelines differ by mechanism, not by effort. Instead, here is the breakdown by what the tactic actually changes.

The pattern is consistent: access changes fast, structure changes at crawl speed, association changes at publication speed, and training does not change on your timetable. So any plan that promises a single timeline across all four is not distinguishing between them.

What does this cost to run in-house?

Here are hours instead of prices, because hours transfer across markets and rates do not.

Overall, a realistic first 90 days for a small team is roughly 40 to 60 hours of setup plus 10 to 15 hours a month of maintenance, with the earned mention work sitting on top of that and scaling with ambition. Tool spend is optional at the start: manual prompt testing costs nothing but time, and the free first-party data sources are Bing Webmaster Tools and Search Console’s generative AI report. If you also want a comparison of the paid options and what they actually measure, we published AI visibility tools for exactly that.

The uncomfortable arithmetic: the earned mention line is both the most expensive and the best evidenced. In contrast, everything cheaper on this table has weaker support.

Which AI visibility metrics should you track: mentions, citations or referrals?

Track all three, but keep them separate, because each one answers a different question. A mention shows the answer named you, while a citation shows it linked your page. A referral, in turn, shows that someone clicked through. Then add accuracy, since a mention with wrong facts still misleads buyers.

Be careful with vendor scores, however. An AI visibility score is one vendor’s own index, so two tools can score the same brand on different scales. For that reason, report the underlying rates instead. Our explainer on AI citations and mentions goes deeper on the first two rows.

How do you prove AI visibility work made money?

Nobody publishes a method for this, so we will be explicit about what is possible and what is not.

For one thing, you cannot do last-click attribution on AI answers. Most AI recommendations produce no click at all, and when they do, referral data is thin. One practitioner puts AI referral traffic under 1% and argues AI search is best understood as a branding channel rather than a traffic channel. Treat it that way, and then the measurement problem becomes tractable.

Which four signals can you actually measure?

Four things you can actually measure:

1. Visibility percentage across a fixed prompt set, run repeatedly, as a percentage rather than a position. The consistency research is clear that any tool reporting a “ranking position in AI” is overstating what the data supports, and equally clear that visibility percentage across dozens to hundreds of prompts run multiple times is a reasonable metric. So fix your prompt set, run it monthly, and watch the percentage.
2. A “how did you hear about us” field on your lead form, with an AI assistant option. It is crude and self-reported, but it is the only direct signal most companies will get. For example, one team measured 3.25% of new users arriving via an AI assistant channel this way.
3. Branded search volume and direct traffic. After all, if AI is recommending you without linking, the downstream effect shows up as people searching your name.
4. Copilot citations from Bing Webmaster Tools, as the one hard number. It is one surface only and Microsoft says it is a count of citations rather than a measure of importance. With Search Console’s generative AI report, it is, in fact, one of only two first-party figures in the discipline.

What to refuse: a report that shows an AI ranking position, a report that shows AI crawler hits as a result, and any ROAS figure attached to AI visibility. The first two measure the wrong thing, while nobody can calculate the third from available data. Our view on what a defensible report looks like is in automated SEO reporting.

What do Search Console, Bing Webmaster Tools and GA4 actually show about AI visibility?

Each free report covers one slice. Bing Webmaster Tools counts citations in Copilot and Bing AI answers, Search Console’s Generative AI performance report counts impressions in AI Overviews and AI Mode, and GA4 counts clicks that arrive from AI assistants such as ChatGPT. None of them shows how often an answer names you without a link, which is why the prompt set still matters.

Which free report shows what?

Here is how each free report compares with your own prompt set.

Where is each of these reports documented?

The sources behind each row are Microsoft’s AI Performance announcement, Google’s Generative AI performance report help page, Google’s Search generative AI control help page and Google’s GA4 default channel group definitions. In short, the two Google reports measure different things: Search Console counts impressions inside Google, while GA4 counts visits that left another AI assistant and reached your site.

Which 2026 changes should you check before you report anything?

Six platform changes from 2026 affect what your baseline means. Check each one once, write down the date you checked, and you will not misread a jump in the numbers later.

• February 10: Bing Webmaster Tools added AI Performance. Any Copilot citation trend you report starts here.
• May 15: Google published a guide to generative AI features stating that seeking inauthentic mentions across the web is not as helpful as it might seem, and that AI text files and special markup are not needed for Google Search.
• May 15: Google clarified that its spam policies also apply to generative AI responses in Search, which puts the riskiest tactics in the trouble table below on the record.
• May 27: Preferred sources began rolling out to AI Mode and AI Overviews, which makes the Preferred Sources step in Tier 1 worth doing now.
• June 3: Google announced the Search generative AI control and generative AI performance insights in Search Console, testing first with a subset of UK sites.
• August 31: Both reached all websites worldwide. Open Settings, then Search generative AI, and confirm the site is set to include. Changes generally take a few days, so log the date of any change before you compare months.

For the tool side of measurement, our comparison of AI visibility tools covers what the paid trackers add on top of these free reports.

How do you check your brand’s AI visibility for free?

You can check your AI visibility for free with a fixed prompt set, a spreadsheet and two first-party reports. As the cost table above shows, each manual round takes 3 to 5 hours. Paid trackers automate the same idea, so start by hand and buy a tool only when the sheet gets too big.

1. Write 20 to 40 prompts in your buyers’ words, as full questions with their real constraints. Google says the average AI Mode search is triple the length of a traditional Search query, so short keyword prompts miss how people ask.
2. Run each prompt three times in a fresh session on each platform you care about.
3. Log every run in the sheet below, including the exact sentence used about you.
4. Pull the free first-party data: Bing Webmaster Tools AI Performance for Copilot citations, and the Search Console Generative AI performance report for Google.
5. Repeat monthly with the same prompts, and then compare quarters rather than single months.

For a quick first look before you build the sheet, our free AI visibility checker shows which AI crawlers your robots.txt blocks and how often AI tools name you against competitors.

What does a free AI visibility tracking sheet look like?

Use one row per run, and add a date column and a competitors column. The rows below are made-up examples for a fictional bakery, not real data.

In this example, prompt P07 names the brand in two of three runs, so its mention rate is 67%. However, three runs are far too few to mean anything, which is why the next section matters. For the full method and a comparison of paid trackers, see our guide on how to track brand visibility in AI search.

How many prompt runs do you need before a change in AI visibility is real?

More than most monthly reports use. At a 30% visibility rate, 60 runs on one platform give a margin of error of about 12 points either way, so a move from 30% to 40% can be pure noise. You need around 300 runs per platform before a month-over-month change of about 7 points means something.

Here is the arithmetic, so you can size your own prompt set. The margin of error for a visibility percentage is 1.96 × √(p × (1 − p) ÷ n), where p is your visibility rate and n is the number of runs on that platform. Comparing two months widens it by about 1.41 times, because both months carry their own error.

What are the limits of this math?

Two honest limits. The formula assumes each run is independent, and repeated runs of the same prompt are not, so your real uncertainty is wider than the table shows. So treat these numbers as the best case. And the margin is widest near 50% visibility, so brands in the middle of the pack need the most runs to see movement.

The practical rules that follow from the table:

• Report quarterly trends from a 20 to 40 prompt set. Then treat single-month moves under about 12 to 16 points as noise.
• Run more prompts rather than more repeats when you can. Wider coverage also lowers the risk of a prompt set that flatters you.
• Also, track each platform separately. Pooling five platforms into one number hides the fact that each one has its own small sample.

Each row of your tracking sheet should record the date, platform, prompt ID, run number, whether the answer named you, the cited URL if any, and the exact sentence used about you. That last column is what turns a percentage into something you can act on. Our guide to ChatGPT rank tracking shows the same sheet applied to one platform.

Who should own AI visibility work inside a company?

One of 33 pages raised this, in a single paragraph. It matters because the work splits across teams that do not normally share a plan, and that is why it stalls.

The last row is the one we would actually insist on. That is because the failure mode we see is not underinvestment. Instead, it is paying for measurement and calling it improvement. If you are evaluating outside help, our guide to choosing a generative engine optimization agency sets out what to ask, and we checked how little the agency shortlists overlap. Measurement is a diagnostic. However, vendors sell it as a growth lever, often alongside the measurement tool they make.

How to improve AI search visibility for enterprise software

Enterprise software buyers ask AI for shortlists, and the shortlist behavior is different from consumer categories in three measurable ways.

Why does category width matter for software brands?

To begin with, category width changes everything. The consistency research found brand visibility varied from 90 to 100% in narrow categories down to 30 to 40% in wide ones, and in one narrow case a single organization hit 97% visibility, appearing in 69 of 71 responses. Enterprise software categories, however, are usually wide. As a result, your realistic ceiling on any single prompt is lower, and your prompt set needs to be wider to detect movement.

In addition, buyers write prompts about needs, not brands. For example, academic work on LLM recommendation found that needs-led and contextual queries change which brands a model retrieves. So a prompt set built from your category name will mislead you. Instead, build it from the problem statements your sales team actually hears.

Finally, product specifics outweigh brand name. In one controlled academic setup, product quality and specification accounted for 82.4% of ranking variance against 1.2% for brand name. For enterprise software that argues for publishing real specifics, integration lists, limits, pricing structure and supported configurations, in extractable form, rather than positioning language. It also argues against thin comparison pages, which is consistent with the listicle collapse above.

Does authority-sounding language help?

One warning from the same body of research. Researchers measured persuasive, authority-sounding language shifting LLM recommendations in the absence of any real quality difference, roughly equivalent to 0.17 rating points of bias. However, the same work found that once every competitor adopts the trick, the individual advantage collapses to near zero. It is documented as a manipulation dynamic in a controlled test, not as a durable tactic, and we are naming it here so you recognize it when someone sells it to you.

How do you make your brand readable to AI shopping and booking agents?

You make your brand readable to AI agents by keeping product and business data complete, current and machine-readable where agents read it. Google’s AI optimization guide describes agents as systems that do tasks for people, such as booking a reservation. It also says protocols like the Universal Commerce Protocol (UCP) will let Search agents do more.

Two platforms already document agent checkout. Google announced UCP on January 11, 2026 as an open standard for agentic commerce, and said it would power checkout on eligible product listings in AI Mode and the Gemini app, starting with eligible US retailers. Meanwhile, OpenAI’s shopping help page says ChatGPT may show an Instant Checkout option for some eligible products and merchants.

Most of this is feed work you may already do for Shopping ads. For the Google side, see our guides to Merchant Center AI attributes and the Business Agent in Merchant Center.

Which of these tactics could actually get your brand in trouble?

No page we read discusses this, including the pages recommending the riskiest tactics. In fact, one openly advises getting mentioned on Reddit by any means necessary, whether as the brand or anonymously, on the same page where it warns against fake mentions.

The general rule we apply: if the tactic only works while the platform does not notice it, it is not a strategy, it is a timing bet.

How to ensure company visibility in AI: the first 30 days

Ordered so each step gives you information the next one needs.

1. Days 1 to 3. Pull server or CDN logs and confirm OAI-SearchBot, PerplexityBot, ClaudeBot, Claude-SearchBot and Bingbot are receiving 200 responses. Then fix any blocks. This is the highest ratio of outcome to effort on the whole list.
2. Days 3 to 5. Claim Bing Webmaster Tools, submit your sitemap, and open the AI Performance report. You now have free first-party citation data for Copilot; open Search Console’s generative AI report for Google at the same time. Then note the baseline.
3. Days 5 to 10. Build a prompt set of 20 to 40 prompts in three buckets: core prompts where you should already appear, competitive prompts where a rival dominates, and experimental prompts testing a new angle. Also, write them from real buyer language, not category names.
4. Days 10 to 12. Run every prompt three times in fresh sessions across ChatGPT, Google AI Mode, Copilot, Perplexity and Claude. Then record appearance as a percentage, not a position. This is your baseline and it will be noisier than you expect.
5. Days 12 to 15. Read which sources the answers cite. In effect, this list is your actual target list. Then note which ones already mention you and which do not.
6. Days 15 to 25. Restructure your three most important pages to answer the core question in the first 40 to 60 words, then support it with citations and statistics. Three pages, not thirty. After all, the Copilot data shows citations concentrate on a handful of pages regardless.
7. Days 25 to 30. Start outreach on the two or three citing sources from step 5 that do not mention you. Expect this to take months, and expect it to produce most of your eventual results.

What should you leave off the first-month list?

What is deliberately not on this list: publishing llms.txt, adding schema as an AI tactic, a content refresh sprint, and buying a tracking tool. Still, none of them is off limits. They are all Tier 3, and they should not come before the seven steps above.

Improve brand visibility in AI search FAQs

How long does it take to improve brand visibility in AI search engines?

It depends entirely on which lever you pull. Unblocking AI crawlers shows up in days. Bing indexation takes one to four weeks. Restructuring a page shows up two to eight weeks after it is recrawled. Earned third-party mentions take two to twelve weeks from publication. Changing what a model knows about you in training data is not schedulable at all. Anyone quoting one number for the whole discipline is not distinguishing between these.

Can you pay to get your brand mentioned in ChatGPT or AI answers?

Not directly. No platform sells placement inside its cited answers today. You can pay for advertising on some AI surfaces, and you can pay for PR that earns a mention on a source the model retrieves from, which is an indirect and legitimate route. Anyone offering to place your brand inside AI answers for a fee is either describing PR in misleading language or describing something you should not buy.

Does schema markup improve AI visibility?

The two controlled tests we found say no. A 1,885 page difference in differences test found no increase on any platform and a small decline on Google AI Overviews. A seven-platform experiment found no measurable influence for most platforms. The one counterweight is Microsoft, whose team has publicly said schema helps its models understand content for Copilot, and neither test isolated Copilot. Add schema for its other benefits, but do not buy it as an AI visibility service with a percentage attached.

Is llms.txt worth publishing?

On current evidence, it is close to free and close to useless for Google. Google has stated it does not use it. A 90-day server log study recorded 84 hits out of 62,100 AI bot visits. Ahrefs found 97% of llms.txt files across 100,000 domains got no requests at all. There is one documented counterexample involving Google AI Mode citing the file after a manual Search Console submission, on a single site with branded German queries. Publish it if it costs you an hour. Do not build a strategy on it and do not pay for it.

Why does my competitor appear in AI answers when I rank higher on Google?

Because ranking and citation are related but not the same. Ahrefs found only about 12% of AI cited URLs ranked in Google’s top 10. One documented page earned roughly 1,200 Microsoft Copilot citations while sitting at Google position 22.4. Copilot reads the Bing index, not Google’s, and ChatGPT leans heavily on third-party mentions and a live search layer. Check your Bing indexation first, then check which sources the AI answer cites and whether you are named on any of them.

How do I check if AI is mentioning my brand without paying for a tool?

Two free methods. Run a fixed set of 20 to 40 buyer-language prompts three times each in fresh sessions across ChatGPT, Google AI Mode, Copilot, Perplexity and Claude, and record the percentage of runs in which you appear. Then claim Bing Webmaster Tools and read the AI Performance report, which gives you an actual count of Copilot citations and the grounding queries that produced them. Between them those two cover diagnosis adequately. We walk through the manual method in detail in our guide to tracking brand mentions in AI search, and the ChatGPT-specific version in ChatGPT rank tracking.

Should I block AI crawlers to protect my content?

Understand what each one does before you decide, because the controls are not interchangeable. Blocking GPTBot keeps you out of OpenAI’s training data but does not remove you from ChatGPT search. Blocking OAI-SearchBot removes you from ChatGPT search entirely. Anthropic splits training, user-triggered fetches and search indexing across three separate crawlers. You can opt out of training and stay in search, and most brands should want exactly that configuration. Check what your security product is doing by default, because many block AI crawlers without anyone deciding to.

Does posting on Reddit or Quora get my brand mentioned by AI?

The evidence for this is weaker than the advice suggests. Research shows AI answers frequently cite Reddit and Quora as domains, which is a descriptive finding. No published test shows that your own posting there increases the odds of your brand being mentioned, which is the prescriptive claim people make from it. Reddit’s share of ChatGPT citations has also collapsed twice, once in late 2025 and again in August 2026, the second time by 86.4% in four days with no published explanation. Treat any platform-dependent tactic as a tenancy, not an asset.

How many prompts do I need to measure AI visibility reliably?

More than most people use. In a study of 2,961 runs across ChatGPT, Claude and Google AI, the chance of getting an identical brand list twice was under 1 in 100. A single prompt run once tells you nothing. Twenty to forty prompts, each run at least three times, gives you a percentage stable enough to compare month over month. Measure percentage of appearances, never position.

Is AI visibility work different for ecommerce?

Yes, in emphasis. Product specifics carry more weight than brand language, product pages have been rising as a share of retrieved pages while listicles and comparison pages fell sharply in August 2026, and feed and catalog data enters the picture in a way it does not for services. We cover the specifics in AI SEO for ecommerce.

Do I need an agency for this?

For the first five steps above, no. Crawler access, Bing Webmaster Tools, prompt set design and manual measurement are all in-house work and we have laid out the hours. Where outside help earns its fee is earned mentions, which is the best evidenced and most expensive part, and in telling you which of the things you are being sold are Tier 3. If you want that read on your current setup, talk to us or see how we approach it as an AI SEO company.

Does GA4 show traffic from ChatGPT and other AI assistants?

Yes. GA4’s default channel group includes an AI Assistant channel that collects referrals from sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok, using the medium ai-assistant. It does not include Google’s AI Overviews or AI Mode, and it cannot see people who read an AI answer and then type your address, so it undercounts AI influence.

How do you tell if your brand is excluded from AI recommendations?

Work from access to answers. Confirm AI crawlers get 200 responses in your logs, check the Search generative AI control in Search Console, and confirm your key pages are indexed in Bing. Then run a fixed prompt set several times on each platform. If competitors are named and you never are, the gap is usually third-party mentions rather than your own pages.

How do you use PR for AI visibility?

Use PR to get named on the sources AI answers already cite. Run your prompt set, list the third-party pages that appear as citations, and pitch those publishers something worth quoting, such as original data or a clear expert view. In the one published 30-day cold-start test, earned placements produced 85.8% of citations against 14% for owned content.

What is an AI visibility score?

An AI visibility score is one vendor’s own index of how often AI answers name or cite your brand across the prompts that tool tracks. Each vendor builds it differently, so two tools can score the same brand on different scales. Use a score to follow one tool over time, and report mention rate, citations and AI referrals alongside it.

How do I get my brand mentioned in ChatGPT answers?

Start with access. Allow OAI-SearchBot and confirm it gets 200 responses, because blocking it removes you from ChatGPT search. Next, run your buyer prompts and list the sources ChatGPT cites. Then earn genuine mentions on those sources. In the one published 30-day cold-start test, earned placements produced 85.8% of citations.

How long does it take to appear in Google AI Overviews?

Google gives no timeline. Its documentation says a page only needs to be indexed and eligible to show as a snippet, with no special markup, so the first wait is crawling and indexing. However, eligibility does not promise a citation. Track the page in the Search Console Generative AI performance report to see when it starts appearing.

Do ChatGPT ads change which brands ChatGPT recommends?

No, according to OpenAI. Its August 31, 2026 update says ads in ChatGPT are always clearly labeled and separate from its answers, and that advertising does not influence the answers ChatGPT provides. Its shopping help page also says product results are not ads. So an ad buys a labeled placement, not a mention inside the answer.

Sources and how we verified them

We read the primary documentation from Google, OpenAI, Perplexity, Anthropic and Microsoft, nine published studies, and practitioner threads on Hacker News, Product Hunt and Indie Hackers. Every number on this page comes from one of those, or from a source named in the sentence where it appears. Where a claim is one person’s single-site account rather than a study, we have said so in the text.

Two limits are worth stating. No Reddit thread or YouTube video is quoted on this page. And we found no verified first-person account of any brand getting an AI platform to correct a false claim, which is why that section says so plainly instead of offering a method.

Platform documentation: Google AI features and your website, Google guide to optimizing for generative AI features, Google Preferred Sources, OpenAI crawler documentation, OpenAI ChatGPT search help, Perplexity crawlers, Anthropic crawler documentation, Anthropic web search tool, Microsoft AI Performance in Bing Webmaster Tools.

Which studies and experiments did we use?

Studies and tests: the generative engine optimization paper, Ahrefs schema markup causal test, Ahrefs AI citation and ranking overlap, Semrush technical SEO and AI search study, Semrush reasoning mode citation instability, Seer Interactive on SearchGPT and Bing overlap, BrightEdge AI Overview and organic overlap, SparkToro research on AI recommendation consistency, brand retrieval and ranking in LLM recommendations, incumbent advantage and brand bias in LLM recommendation.

Experiments and practitioner reports: the llms.txt server log experiment, the schema markup multi-platform experiment, the 2,000 AI blogs deindexation experiment, the cats.txt falsifiability test, the llms.txt counterexample with access logs, GEO experiments challenging conventional advice, the listicle and comparison page citation collapse, the Reddit citation collapse mechanism, the second Reddit collapse, ranking and AI citation correlation across 400 keywords, Bing AI Performance first-party analysis, a single operator’s Copilot citation account, an agency’s 90-day Copilot citation data, schema markup in AI search without the hype, Google on normal SEO and llms.txt, practitioners on whether GEO is a new discipline, on generative information retrieval misinformation, why AI cannot be tracked like traditional search.

Hustle Marketers is a Google Partner, Meta Business Partner and Microsoft Advertising Partner, rated 4.9 out of 5 from 591 reviews, with a Clutch rating of 5.0 from more than 50 reviews and 6 Clutch Awards in 2026. Ishant Sharma, who founded the agency, has worked in search since 2013. We have run SEO and paid media for more than 2,500 brands.