Technical SEO Checklist 2027: 105 Checks for AI Search

Most technical SEO checklists still audit for one reader. That reader is Googlebot. But your pages now get judged by a second group as well, and that group behaves nothing like a search crawler. For example, GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot arrive with less patience, weaker rendering and different goals. Some of them never arrive at all, yet still quote you. This checklist covers all 105 checks we run on client sites, split across 18 categories. Every item tells you why it matters and how to check it, so you can hand the list straight to a developer. For the strategy sitting on top of these fixes, see what Google documents about AI search and what it stays silent on.

The shift nobody has priced in yet is this: AI systems are not simply crawlers you allow or block, they are users who have to read your page, render it, and increasingly act on it.

Updated for 2027: we rechecked this list against Google and Microsoft documentation changes published through October 2026. The count is still 105 checks in 18 categories. What moved is in the new section on what changed in technical SEO for 2027, including the new Search Console control and report for AI Overviews and AI Mode. This edition also corrects how Google treats 404 and 410 responses and what blocking ClaudeBot does, and it adds hreflang checks, platform notes and a symptom-to-check troubleshooting table.

Key Takeaways

• 105 checks, 18 categories, 4 priority levels. Every item states why it matters and exactly how to check it, and the first 75 are infrastructure checks while the last 30 cover keyword, content and link diagnostics.
• Fix these four first: the four-variant redirect test on a deep URL, the filetype: search for exposed documents, each AI user agent line in robots.txt, and your crawled-but-not-indexed ratio. Under an hour combined, so start there.
• Ten things here appear in no other published technical SEO checklist we audited, including the log file method for finding rendering gaps, accessibility tree testing, agent commerce readiness, and why you get cited by bots that never visit you.
• llms.txt is not robots.txt for AI. Three widely-shared checklists get this wrong, while Google now says Google Search ignores the file. Add it if you want, but do not count it as strategy.
• Do not noindex thin pages. Noindex still leaves them crawled and still counted in sitewide quality. Instead, remove them or improve them.
• Short on time? The 50-item version for small sites is near the end, and it still covers roughly 90% of the risk.

What Is a Technical SEO Checklist and Who Needs One

A technical SEO checklist is a repeatable list of infrastructure checks that confirm search engines and AI systems can reach, read, render and trust your website. It sits underneath content and links. When it breaks, nothing above it works.

You need one in four situations. First, before a migration or replatform. Second, when organic traffic drops without an obvious content cause. Third, when you inherit a site and have no history on it. Fourth, on a fixed cadence, because technical debt accumulates quietly as a site grows.

In general, run the full 105 checks quarterly on a large site. On a small brochure site, however, run the 50-item short version twice a year. Then use SEO automation tools to monitor the critical items monthly.

How This Checklist Is Prioritised

Every check carries one of four labels. So treat them as a work order, not a suggestion.

Technical SEO vs On-Page SEO: Where Does Each Check Belong?

Technical SEO covers whether search engines and AI systems can reach, render, index and trust your pages, while on-page SEO covers what each page says and targets. In this checklist, checks 1 to 75 are technical infrastructure, and checks 76 to 105 are the keyword, content and link diagnostics that sit on top of it.

We keep all 105 in one list because an audit fails when two teams run the halves and never compare notes. For example, a page can pass every crawl check and still miss, simply because it targets the wrong intent. So the table below shows where each group sits and who usually fixes it.

That split also explains the count: the title says 105 because every numbered row on this page is a check you can pass or fail, and 75 of them are strictly technical. If you only need the infrastructure audit, stop after check 75. However, when a page is indexed but still not winning, the last 30 usually tell you why. For a shorter view of what Google itself requires, see our page on technical SEO requirements.

What Changed in Technical SEO for 2027?

Three things changed how this checklist runs going into 2027. Google added a site-wide Search Console control and a performance report for AI Overviews and AI Mode, Google published guidance that rules out several popular AI tactics, and Microsoft added Copilot citation reporting to Bing Webmaster Tools. However, none of them adds a new crawl or index requirement. Instead, they change how you run checks you already have.

Here are the dated changes that matter most, as well as the checks on this list each one touches. After that, a second table lists the smaller documentation changes.

What Changed for Spam, AI Controls and Reporting?

The full list of Google’s dated changes is on the Search Central documentation updates page, which is worth a monthly look if you run audits for a living.

Which of These Changes Should You Add to Your Audit First?

The June 3 row is the one to add to your audit today. After all, a site-wide switch that removes you from AI Overviews and AI Mode is the kind of setting one person changes once and everyone forgets. Google says the control is not used as a ranking or inclusion signal for other parts of Search, and that changes generally take a few days to apply, so record its state every time you run check 25.

Similarly, the reporting rows matter for the same reason. Search Console’s Generative AI performance report shows impressions only, for AI Overviews and AI Mode combined, with no clicks or queries, and it may not appear at all on a site with too few AI impressions. So your AI visibility picture still has to be built from several sources. Our guide on how to improve brand visibility in AI search shows how to combine them, and tracking brand mentions in AI search covers the manual prompt method.

Which Other Google Documentation Changes Touch These Checks?

Smaller documentation changes matter too, because each one shifts how a specific check runs. Google’s Search Central changelog and blog list these (checked October 29, 2026).

Setup and Access: The Three Checks That Come First

You cannot audit what you cannot measure. So before anything else, confirm you actually own the data.

Also add Bing Webmaster Tools to that list. It costs ten minutes and it matters more than it used to, because ChatGPT leans on the Bing index for live answers. We cover that properly in our guide to ranking in ChatGPT search.

Crawlability: 12 Checks Before Anything Else Counts

Crawlability is the first prerequisite. After all, without a crawl there is no index, and without an index there is no ranking and no AI citation. So this section runs before every other section in a real audit.

How Do You Catch Crawl Waste and Broken Pages?

How Far Crawl Budget Can Actually Move

Crawl budget gets treated as a fixed constraint. In fact, it is not. For example, in one case reported by Jairo Guerrero of AirOps, a client site went from roughly 8,000 crawl requests per day to about 1.2 million. As a result, a full-site crawl fell from an estimated thirteen years to around a month.

That is a single practitioner account rather than a published study, so treat the exact figures as directional. The mechanism behind it, however, is not in dispute. Crawl capacity responds to server speed, internal link structure, response codes and the ratio of useful pages to junk. All four are things you control, so the ceiling moves when you fix them.

Then track the number in the Search Console Crawl Stats report month over month. A flat line while you publish is a warning. A falling line while your URL count grows is the shape that usually precedes a ranking drop. Still, most sites under 10,000 pages never hit a crawl budget limit, as the FAQ below explains.

Why 410 Beats 404 for Injected URLs

This one deserves more than a table row, because almost every other checklist recommends robots.txt or noindex for parameter junk. That advice fails when the URLs were never yours.

A negative SEO attack or a compromised plugin can generate hundreds of thousands of fabricated URLs overnight. In one case we reviewed, a site went from roughly 10,000 known pages to over 500,000 discovered URLs, and rankings fell within days. However, blocking those URLs in robots.txt makes it worse. Google can no longer crawl them, so it cannot see that they are gone, and they linger in the index, which matches Google’s crawl budget guide (checked October 29, 2026): blocked URLs stay in the crawl queue much longer.

What Should Junk URLs Return Instead of a robots.txt Block?

Serve 410 Gone instead, because a 404 says the page might come back while a 410 says it will not. However, Google treats both the same way, and its HTTP status code guide (checked October 29, 2026) says every 4xx code except 429 gets the URL removed from the index. So 410 beats 404 on clarity, in your logs and for the next developer, not on speed. Match the injected pattern at server level rather than page by page, because you cannot hand-edit half a million URLs.

Blocking junk URLs in robots.txt is the most common wrong answer in a spam cleanup. You are not hiding the problem from Google, you are hiding the solution.

Indexation: 8 Checks That Decide If You Exist

Crawling gets you read. Indexing then gets you eligible. And because AI systems retrieve from search indexes first, an unindexed page cannot be cited even by a chatbot that has never visited your server.

How to Read Your Crawled-But-Not-Indexed Ratio

Most guides tell you to open the GSC Page indexing report. Very few tell you what the number means, or why automated reports never match Search Console when you cross-check it. So here is the version we use with clients.

To begin with, take indexed pages and divide by total known pages. Then band the result.

The reason the low band is a quality problem rather than a crawl problem is worth spelling out. For instance, Guerrero describes a client where roughly 80 percent of crawled pages carried no value at all, and the effect was not confined to those pages. As a result, the junk dragged down the whole-site assessment, including the pages that were fine.

The instinct at that point is to noindex the weak pages. Resist it. Noindex removes the page from results but Google still crawls it and still weighs it when judging the site. Instead, deleting or genuinely improving the page is what changes the sitewide picture. We go deeper on this in our programmatic SEO guide, where the problem shows up fastest.

Robots and Directives: 3 Checks With Outsized Consequences

In fact, one wrong character in robots.txt can switch off an entire site. This is the shortest section in the checklist, yet also the most dangerous.

The AI User Agents Worth Naming Individually

Blocking AI crawlers is a real business decision, not a default. But most sites block them by accident, through an inherited rule or a security plugin. So check each agent separately, because they do different jobs.

Note the split in that last row. Google-Extended is not Googlebot. Blocking it does not affect your Google Search rankings, which is exactly why so many teams get the decision wrong in both directions.

Which AI Crawlers Can You Actually Block With robots.txt?

You can block the training and search crawlers with robots.txt, but the fetchers that act on a user’s request may not obey it. OpenAI says robots.txt rules may not apply to ChatGPT-User, and Perplexity says Perplexity-User generally ignores them, because a person asked for the page (both checked October 29, 2026).

Anthropic works differently. Its help center (April 2026) lists three agents, and it says blocking Claude-User stops Claude from fetching your pages for user questions. So check every agent you care about, not just the famous ones.

Both OpenAI and Perplexity say a robots.txt change can take about 24 hours to reach their systems. In addition, robots.txt is only half the picture. If your firewall blocks an operator’s published IP ranges, the bot fails even when robots.txt allows it, so review check 48 alongside check 25.

For Google, blocking Google-Extended does not remove you from AI Overviews or AI Mode. Instead, Google’s AI features guide (checked October 29, 2026) names robots.txt rules for Googlebot and snippet controls as the way to manage those features, alongside the Search generative AI control in Search Console.

A Straight Answer on llms.txt

Several checklists now recommend llms.txt as if it were robots.txt for AI. That framing is wrong, and it is worth saying plainly.

llms.txt is a proposed plain-text file listing your key pages for language models. No major AI provider has committed to reading it. Google’s guide to optimizing for generative AI features also says, in a note added on June 15, 2026, that Google Search ignores these files, so they neither help nor hurt your visibility there. More importantly, it does not match how these systems actually work, because they retrieve through search indexes rather than browsing a directory of your choosing.

So add it if you like. After all, it takes ten minutes and costs nothing. Just do not count it as an AI visibility strategy, and do not let it displace the checks in this list that genuinely move the needle.

Duplicate Content and Canonicals: 5 Checks

Duplicate content rarely arrives as a copy-paste problem in the first place. It arrives as an architecture problem, and the fix is usually a redirect rule rather than a rewrite.

The Four-Variant Test Almost Nobody Runs Properly

Here is the check itself, and it takes two minutes. First of all, pick a deep URL, not your home page. For example, a product page or a blog post is ideal. Then request all four of these and record the response.

• http://domain.com/deep-page/
• https://domain.com/deep-page/
• http://www.domain.com/deep-page/
• https://www.domain.com/deep-page/

Three of them should return a single 301 to the fourth. If any of them returns a 200, then you have a live duplicate. Similarly, if any of them returns a chain of two or more redirects, you are leaking crawl budget on every request.

We test the deep URL rather than the home page because home-page-only rules are the single most common miss. A server rule that redirects the root correctly can still pass inner paths through untouched.

How Do You Check Hreflang and International SEO Setup?

Hreflang checks only apply when you publish the same page in more than one language or for more than one country, which is why they sit outside the 105 numbered checks. If that is your site, add the five checks below to the duplicate content pass, because hreflang is how Google learns which version to show in each market.

Also note that Google treats localized pages as duplicates only when the main content is left untranslated. So a translated template with untranslated body copy is a duplicate content problem, not an hreflang problem.

Site Architecture and Internal Linking: 8 Checks

Internal links are how discovery actually happens, and they are the cheapest ranking lever most sites still ignore.

To see which links on a page are followed, including a page-level meta robots nofollow, run it through our free dofollow link checker.

Rendering and JavaScript: The Section Most Checklists Skip

Googlebot renders JavaScript. However, most AI crawlers do not, or they give up far sooner. So a page that looks complete in your browser can reach an AI system as an empty shell. That is why this layer gets two checks of its own, numbers 40 and 41.

The Two-Minute Raw HTML Test

Open the page, then use view-source rather than inspect element. View-source shows what the server sent. Inspect, by comparison, shows what your browser built afterwards.

Then search that raw HTML for your H1, your first paragraph, your price and your primary call to action. If any of them is missing from view-source but visible on screen, JavaScript is producing it. Googlebot will probably still see it. An AI crawler, however, often will not.

The Log File Check That Finds Rendering Gaps

This is the check we have not seen in any competing checklist, and it is the one that finds problems the other 104 checks miss.

When a bot fetches a page, it requests the HTML document. If it then also renders, it requests the JavaScript and CSS assets that page depends on. So you can compare the two counts directly in your server logs.

To begin with, filter your logs by user agent. Count HTML document requests. Then count requests for the JS bundles those pages load. If a bot fetched a page template 10,000 times but pulled the associated JavaScript only a few hundred times, that bot is not rendering your pages. As a result, whatever your JavaScript produces is invisible to it.

So that single ratio tells you whether server-side rendering is a nice-to-have or an emergency. And it is measurable today, because you already have the logs.

Speed and Core Web Vitals: 4 Checks

Speed stopped being purely a ranking conversation. It is now also a citation conversation, because a crawler on a budget abandons slow responses.

Does Site Speed Affect AI Citations?

There is emerging evidence that this now affects citations as well as rankings. A study of more than 2,000 sites, cited in the AirOps session, reported that faster sites were referenced roughly 20 to 30 percent more often in AI answers, with one slide putting the gap at 1.47 times. We have not reviewed the underlying methodology, so treat those numbers as a direction of travel rather than a benchmark.

The mechanism is intuitive enough. In short, a crawler working through a large candidate set on a fixed budget abandons slow responses. Googlebot usually comes back later. In contrast, an AI crawler often does not.

Google’s published Core Web Vitals thresholds are LCP under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1. Measure against field data where you have it, because lab scores flatter you. web.dev adds (checked October 29, 2026) that a page passes only when it meets all three targets at the 75th percentile of page loads, split by mobile and desktop.

How Many Sites Pass Core Web Vitals Today?

Fewer than half of mobile sites pass. The HTTP Archive Web Almanac 2025, published in January 2026, found that 48% of mobile websites and 56% of desktop websites had good Core Web Vitals in its July 2025 data, up from 44% and 55% in 2024.

So a pass is still an edge on mobile. Also, INP replaced First Input Delay as a Core Web Vital on March 12, 2024, according to web.dev, so any audit template that still grades FID is out of date. In that case, swap the FID column for INP before you run checks 44 and 45. Our Core Web Vitals and INP guide covers the fixes for WordPress and Shopify themes.

Server and Security: 4 Checks That Are Really Business Risk

This section is short, and every item in it is Critical. That is not padding. Each one can take a site off the map in a week, so treat them as emergencies.

Run the Exposed File Search Before Your Client Does

Check 49 is the one that changes the temperature of a client meeting. So run these four searches on any site you take over.

• site:domain.com filetype:pdf
• site:domain.com filetype:xls OR filetype:xlsx
• site:domain.com filetype:doc OR filetype:docx
• site:domain.com filetype:csv

Most of the time you find old brochures. Occasionally, though, you find an internal payroll sheet or a client contact export sitting in an uploads folder. Fix it by removing the file and serving 410, then requesting removal in GSC. Do not simply noindex it, because the file itself remains publicly downloadable.

Mobile: 1 Check, Still Non-Negotiable

On-Page Technical: 5 Checks at Template Level

These are on-page items, but you audit them technically. In other words, you are looking for template errors that affect hundreds of pages at once, not for one weak title.

Structured Data: 6 Checks After Google Removed FAQ Rich Results

Structured data is the one area where a 2027 checklist has to be more careful than a 2023 one. Google removed FAQ rich results from Search on 7 May 2026. HowTo went the same way in 2023. Plenty of checklists still tell you to add both. Six checks below, then two things worth saying plainly.

The Schema Types That Stopped Paying in 2026

FAQPage and HowTo are two of the most commonly recommended types in technical SEO checklists. However, neither produces a result in Google Search any more.

The wider point is that markup follows Google’s feature list, not schema.org’s vocabulary. Schema.org publishes hundreds of types. In contrast, Google renders a couple of dozen. So build for the second list.

In addition, the same clean-up hit other types. Google’s documentation updates page (checked October 29, 2026) shows course info, estimated salary, learning video, special announcement and vehicle listing documentation removed on September 9, 2025, and practice problem documentation removed on January 6, 2026. Google also clarified in November 2025 that Dataset markup is used only by Dataset Search, not Google Search.

A Straight Answer on Schema and AI Search

You will read everywhere that schema is how you get cited by AI. Google’s own documentation on AI features says the opposite in one line: there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode.

What Google does ask for is that the page is indexed, eligible to be shown with a snippet, and that its structured data matches the visible text. That last part is the useful instruction. Schema does not buy you a citation. However, inaccurate schema can cost you one.

The studies reporting large citation lifts from schema are correlational. In other words, sites that mark up properly tend to be sites that do everything else properly too. Mark up for the Search features you are genuinely eligible for, keep it accurate, and treat any AI benefit as a side effect rather than the business case. Our restaurant schema markup guide shows what that looks like for one business type, with Google’s support status for each restaurant feature.

Logs and Monitoring: 4 Checks Nobody Else Teaches

Search Console reports only on Google. So your server logs are currently the only reliable source of truth on AI crawler behaviour. As a result, this section is the highest-leverage part of a modern audit.

What to Do When Your Platform Will Not Give You Logs

Hosted ecommerce platforms are the usual problem here. Shopify, for example, does not expose raw server logs, so you cannot confirm whether an AI crawler ever reached a product page.

Even so, you still have options. Three workarounds get you most of the way.

1. Put the site behind a CDN such as Cloudflare and read the CDN logs instead. The request still passes through it.
2. Use GSC Crawl Stats as a proxy for Googlebot behaviour, then treat it as a floor rather than a full picture.
3. Finally, add lightweight server-side logging through an app or middleware layer where the platform allows it.

If none of those is possible, then factor it into your platform decision. Losing log visibility is a real cost, and it is worth naming at the point of choosing a stack rather than two years later.

AI Search, GEO and AEO: 7 Checks

This is where a 2027 checklist earns its title. In fact, these items are not content advice, they are infrastructure decisions about how retrieval works.

How Retrieval Picks Candidates Before It Picks an Answer

Check 66 tells you to cover topics and entities rather than exact keywords. The reason sits in how retrieval works, and it runs in two stages. Our guide to generative engine optimization covers what to do with that on the content side.

First the system finds candidates by meaning rather than by string match. In effect, your query becomes a numerical representation, and passages that sit close to it in meaning get pulled in. That is why a page can be retrieved for a phrasing it never literally contains.

Then it reranks that pool before choosing what to quote. Guerrero puts the size of that shortlist in the region of 150 results. We have not found published documentation confirming that specific figure, so treat it as a practitioner estimate. The two-stage shape, retrieve broadly then rerank, is well established regardless of the exact number.

The practical instruction is the same either way. Name the entities. Then list the tools, standards, competitors, regulations and adjacent concepts a knowledgeable person would expect. Breadth of coverage gets you into the candidate pool. In contrast, repeating one phrase does not.

Why You Can Get Cited by a Bot That Never Visited You

This is the most misunderstood mechanic in AI search, and it changes what you should optimise.

When someone asks an AI assistant a question, the system does not usually crawl the open web in that moment. Instead, it queries a search index, gets back a set of results, and each result arrives as a title plus a short text extract. If that extract already answers the prompt, then the model can write its answer and cite you without ever requesting your page.

Two consequences follow. First, your server logs will under-report AI visibility, because many citations leave no trace there. Second, the snippet is the artefact being judged, not your full page.

So the practical move is to make the first 200 characters of the relevant section carry the answer. Front-load the fact. Put the number, the definition or the direct response near the top of the passage rather than after three paragraphs of setup.

The scale of the click compression is worth naming, even though the numbers move around by study. Eric Siu cites work putting the organic traffic reduction from AI Overviews at more than 37 percent on affected queries. Published figures in this area vary widely, so use it to size the risk rather than to forecast your own result. Segmenting your own Search Console data by intent gives a far more useful answer than any industry average.

We got this wrong on our own site for most of 2025. We optimised full pages for depth and buried every direct answer under context. Impressions climbed, citations did not.

How Much Do AI Overviews Cut Clicks to Classic Results?

By a lot on informational queries, although the exact share varies by study. Ahrefs found that an AI Overview correlated with a 58% lower average click-through rate for the top-ranking page in December 2025 data, up from 34.5% in its April 2025 study. Pew Research Center found that users who saw an AI summary clicked a traditional result in 8% of visits, against 15% without one.

Questions trigger these summaries most. In the same Pew data from March 2025, 60% of searches that began with who, what, when or why produced an AI summary, compared with 18% of all searches. As a result, question-led informational pages feel the cut first, which is why checks 67 and 70 ask you to segment clicks by intent.

The audience is large, too. On its Q2 2025 earnings call, as reported by TechCrunch, Google said AI Overviews had 2 billion monthly users and AI Mode more than 100 million in the US and India. However, Google’s AI features guide (checked October 29, 2026) says clicks from results pages with AI Overviews are higher quality, so track engaged sessions and conversions next to clicks before you judge a drop.

Agent Readability: 3 Checks for the Web That Clicks Back

AI agents are starting to complete tasks rather than just summarise pages. That means your site now needs to be operable by software, not only readable by it. No other technical SEO checklist we audited covers this at all.

How to Read Your Own Accessibility Tree

The check takes about five minutes and needs no tools beyond Chrome.

1. Open the page and press F12 to launch DevTools.
2. Go to the Elements panel, then open the Accessibility pane on the right.
3. Then enable the full-page accessibility tree view.
4. Find your primary action, for example Add to Cart or Request a Quote.
5. Finally, read its role and its accessible name.

You want to see a role of button or link with a clear name. What you often see instead is a generic node with no name at all. That happens when the element is built like this:

<div class="btn" onclick="submitOrder()">Add to Cart</div>

Visually that is a button. Structurally, however, it is a rectangle. An agent reading the accessibility tree sees no interactive control, so it cannot complete the task. The fix is genuinely this simple:

<button type="submit">Add to Cart</button>

Same appearance with the right CSS, completely different machine meaning. Run this on your cart, your quote form and your booking flow first, because those are the pages where failure costs money.

What Does Google Say About AI Agents Using Your Site?

Google now gives the same advice as this section. Its guide to optimizing for generative AI features (checked October 29, 2026) says browser agents may analyze screenshots, inspect the DOM structure and interpret the accessibility tree when they complete tasks such as booking or comparing products.

The guide also names the Universal Commerce Protocol as one of the emerging protocols that will let Search agents do more, which is the kind of standard check 75 asks you to track. In addition, web.dev’s guide to agent-friendly websites (April 2026) lists screenshots, raw HTML and the accessibility tree as the three main ways agents read a page, and it calls WebMCP a proposed standard that is still in active development.

Keyword and Intent: 13 Checks

Technical work gets you eligible. Intent work, in turn, decides whether eligibility turns into traffic. These checks belong in a technical audit because they are diagnostic, not creative.

Which Checks Cover Targeting, Difficulty and Position Bands?

Why the Page Above You Has Already Done Your Research

Check 82 is the cheapest research method on this list, and the scale of it surprises people. Ahrefs analysis of roughly 3 million searches found that the page ranking first typically also ranks for close to 1,000 other keywords.

So the page beating you has already mapped the topic. Then pull its full keyword profile, filter to positions 1 to 5, and read the result as a coverage map. You are not copying the page. Instead, you are reading the evidence of what the topic actually contains.

The same body of work also sits behind check 55. A study of around 2 million keywords found shorter URLs correlate with better positions, which is why the slug should carry the primary keyword and nothing else. Similarly, check 39 comes from correlation work suggesting pages that link out to high authority resources tend to rank better, not worse.

The Position Band Framework

Most guides stop at low-hanging fruit and mean positions 2 to 10. We widen the first band and add two more, because SERP features and AI answers have pushed the click cliff further up the page.

Content Quality: 10 Checks

Quality is a technical concern because it is measured in bulk. So a hundred weak pages will drag down the twenty good ones sitting beside them.

How Do You Check Readability and Page Polish?

Two Content Quality Numbers Worth Holding On To

Check 94 exists because reading level is a distribution problem, not a matter of taste. Around half the adult population in the United States reads below an eighth grade level. A page written at graduate reading level is therefore not more authoritative. Instead, it is less accessible to half of its audience.

Check 95 is the one most teams skip when they start generating with AI. Eric Siu’s framing is that output reaches roughly 95 percent of usable quality when the model has learned your house style first, against generic output when it has not.

The method is straightforward. Have the model read ten of your strongest existing posts and extract the patterns: typical length, sentence rhythm, how you use tables, how you cite data, where links sit, how you open and close. After that, save it as a standing instruction. Then generate.

Off-Page: 7 Checks

Links are still the clearest external trust signal, and they now feed AI citation patterns as well as rankings.

Full Comparison: What Every Major Technical SEO Checklist Covers

Before writing this, we audited fourteen published technical SEO checklists, including the ones from Semrush, Ahrefs, Backlinko, DebugBear, Sitebulb, seoClarity, Yotpo and Page One Power. We logged what each one covers and, more usefully, what none of them do.

Here is the whole comparison. Everything below reflects those pages as published in August 2026, and every one of them is a genuinely useful resource. The point is not that they are weak. Rather, the point is that the field has a shared gap.

The Ten Things No Other Checklist Explains

Four of the columns above are empty across all fourteen competitors. Five more items are covered so thinly elsewhere that they may as well be. Here is the full list, with an honest note on who else touches each one.

What Do the Ten Gaps Have in Common?

The pattern underneath all ten is the same. Every other checklist treats an AI system as a crawler you either permit or block. None of them treats it as a user that has to read the page, render it, and increasingly act on it.

The 50-Item Short Version for Small Sites

Running all 105 checks on a twelve-page site is a poor use of a morning. So here is the reduced set. Run these, and skip the rest until the site grows past a few hundred URLs.

First of all, keep every Critical item: 1, 4, 16, 24, 25, 27, 46, 47, 48, 49, 56, 57, 76 and 77. Then add these High items: 2, 5, 6, 8, 9, 10, 17, 18, 19, 20, 26, 28, 29, 30, 32, 33, 34, 35, 40, 41, 42, 43, 50, 51, 62, 63, 66, 67, 68, 69, 70, 73, 74, 78, 79 and 89.

That is 50 checks, and it covers roughly 90% of the risk on a small site.

How Do You Run a Technical SEO Audit Step by Step?

Run it in nine passes, in the order the data depends on itself: access first, then quick tests, the crawl, the index, rendering and logs, performance and security, templates, the AI and agent layer, and finally intent, content and links. The numbering on this page is a category order, so use these passes as the work order.

Which Checks Belong in Each Audit Pass?

Every one of the 105 checks sits in exactly one pass, so no check is run twice.

How Should You Prioritize What the Audit Finds?

Then sort every finding by its priority label, not by pass. Across the list, the labels split like this:

Two patterns are worth knowing before you start. All four Server and Security checks are Critical, so pass 6 can produce emergencies even on a site that looks healthy. And none of the seven Off-Page checks is above Medium, so link work never jumps the queue ahead of access, indexing and rendering. If you want this run for you, the ranges and what drives them are in our breakdown of what an SEO audit costs.

What Are the Most Common Technical SEO Issues and How Do You Fix Them?

Most technical SEO problems show up as one of a few symptoms, from pages missing in Google to content AI tools never quote. So start from the symptom you see, then run only the checks that diagnose it.

How Do You Confirm a Fix Worked?

Rerun the same checks after the fix ships. For example, a robots.txt or noindex fix shows up in the URL Inspection tool on the next crawl, while a Core Web Vitals fix needs the full 28-day field window before the report changes. If the symptom stays after that, move to the next likely cause in the row.

Which Tools Do You Need for Each Technical SEO Check?

Google Search Console is the most used tool on this list: it appears in the how-to of 17 of the 105 checks. A site crawler comes next at 14, then keyword and backlink tools and Google search operators at 12 each. And 9 of the 14 Critical checks need nothing more than Search Console, the Rich Results Test, search operators or a plain browser.

We counted every tool named in the How to check column of the tables above. A check can name more than one tool, so the counts overlap.

Which Critical Checks Need Paid Tools?

The five Critical checks that need more than free tools are 4 (a crawler), 46 (a malware scan), 47 (an uptime monitor), 48 (a firewall with bot rules) and 77 (a keyword tool positions report). In contrast, everything else on the Critical list can be done today with what Google gives you for free. To validate markup quickly while you work through pass 7, use our free schema checker.

Bing Webmaster Tools only appears twice, but do not read that as low importance. After all, it is the only free first-party source of Copilot citation data, which is why it sits in pass 1 next to Search Console.

Which Checks Change on Shopify, WooCommerce and Magento?

The 105 checks stay the same on every platform, but the weak spots move. Shopify limits what you can see, while WooCommerce and Magento give you more control and more ways to create duplicate URLs. So run the full list, then spend extra time on the checks below for your platform.

How Often to Run a Technical SEO Audit

Cadence depends on site size and on how much changes between audits. If you are budgeting for one rather than running it yourself, our breakdown of what an SEO audit costs covers the ranges and what drives them.

Where the Data in This Checklist Comes From

Most technical SEO checklists cite nothing. In fact, of the fourteen we audited, four of the highest-authority pages carry no external statistics at all. So it is worth being explicit about where the numbers on this page came from.

The field observations quoted throughout are drawn from five practitioner sessions, each linked below. Where a figure is a single practitioner’s reported client result rather than a published study, we have said so in the text rather than dressing it up as research.

Google’s own documentation is the source for the Core Web Vitals thresholds, how Google treats 404 and 410 responses and the sitemap size limits, while the AI crawler rules come from OpenAI, Anthropic and Perplexity documentation.

Frequently Asked Questions

What is the difference between a technical SEO checklist and a technical SEO audit?

The checklist is the list of things to check, the audit is the act of checking them on a specific site. A checklist is reusable and generic. An audit produces findings, priorities and owners for one property at one point in time. Use this checklist as the input to your audit template.

How long does a technical SEO audit take?

Between four hours and three weeks, depending almost entirely on site size and log access. A small brochure site running the 50-item version takes a focused morning. A large ecommerce site with faceted navigation, multiple templates and log file analysis realistically takes two to three weeks including the write-up.

Do I need to fix every item on the checklist?

No, and trying to is the fastest way to fix nothing. Clear every Critical item first, because those block crawling, indexing or revenue outright. Then work through High items over the following month. Medium and Low items are batch work you schedule, not emergencies.

Should I block AI crawlers like GPTBot in robots.txt?

Only if you have a specific commercial reason, and even then, block selectively. GPTBot collects training data while OAI-SearchBot powers live answers. Blocking the second one removes you from ChatGPT citations entirely. Most businesses that want AI visibility should allow the retrieval agents and decide separately about the training ones.

Is llms.txt worth adding to my site?

It is cheap to add and currently unproven, so treat it as optional housekeeping rather than strategy. No major AI provider has committed to reading it, and it does not reflect how retrieval actually works. Add it if you want, then spend your real effort on indexation, snippets and rendering.

How do I know if AI systems can actually read my page?

Use view-source rather than inspect element, then check whether your key content is present in the raw HTML. If your headline, price or main answer only appears after JavaScript runs, most AI crawlers will miss it. Follow that with a server log check comparing HTML fetches against JavaScript loads for each named AI agent.

Why do my server logs show no AI crawler visits even though I get cited?

Because many citations never require a visit. AI systems query a search index first and receive a title plus a short snippet for each result. When that snippet answers the prompt, the model writes its answer without fetching your page. Your logs will therefore always under-report AI visibility.

Can an AI agent complete a purchase on my website?

Only if your interactive elements are real semantic HTML. Open Chrome DevTools, view the accessibility tree, and check whether your primary action has a button or link role with a clear accessible name. A styled div reads as an inert rectangle to an agent, so the flow fails silently.

What is the most commonly missed check on this list?

Confirming that the http to https and www redirect applies site-wide rather than only on the home page. We find this on a meaningful share of audits. The root redirects perfectly, every inner URL stays reachable on all four variants, and the whole site quietly runs as duplicates.

Does schema markup help you get cited in AI search?

Not directly. Google’s own documentation on AI features states there is no special schema.org structured data you need to add to appear in AI Overviews or AI Mode. What it does ask is that the page is indexed, eligible to be shown with a snippet, and that its structured data matches the visible text. Accurate schema will not win you a citation on its own, but inaccurate schema can cost you one.

Should I remove FAQ schema now that Google has dropped FAQ rich results?

No need to rip it out. Google stopped showing FAQ rich results on 7 May 2026 and confirmed the markup causes no problems if it stays. Just stop treating it as a deliverable, remove it from new templates, and stop reporting FAQ impressions that no longer exist.

Can you opt your site out of AI Overviews and AI Mode?

Yes. Search Console has a Search generative AI control, under Settings, that excludes your site’s links and content from AI Overviews, AI Mode and generative AI features in Discover. Google says it is not used as a ranking or inclusion signal for other parts of Search, and changes generally take a few days. Page-level controls such as nosnippet, data-nosnippet, max-snippet and noindex also limit what Google shows.

Do you need to break content into small chunks for AI search?

No. Google’s guide to optimizing for generative AI features, added in May 2026, says there is no requirement to break content into tiny pieces for AI to understand it, and that there is no ideal page length. Write sections that answer their heading directly, because that helps readers and snippets, but do not split pages into fragments for AI.

What should a technical SEO checklist include?

A technical SEO checklist should cover setup and access, crawlability, indexation, robots.txt, canonicals, site architecture, rendering, speed, security, mobile, template-level on-page elements, structured data and server logs. In 2027 it also needs AI crawler access, AI search visibility and agent readability. This one adds keyword, content and link diagnostics, for 105 checks in total.

Is a 410 better than a 404 for SEO?

Not for Google rankings. Google’s HTTP status code guide (checked October 29, 2026) treats every 4xx code except 429 the same way, and Google’s John Mueller said in 2024 that the difference is too small to matter for SEO. Still, use 410 when a removal is deliberate, because it reads more clearly in logs, and never block junk URLs in robots.txt instead.

Do you need hreflang checks in a technical SEO audit?

Only if your site has versions of the same page for different languages or countries. In that case, check that every version lists itself and all the others, that each pair links both ways, that URLs are fully qualified and that an x-default page exists for unmatched visitors. Otherwise, single-language sites can skip hreflang entirely.

Does blocking Google-Extended remove you from AI Overviews?

No. Google says Google-Extended controls whether your content is used to train Gemini models and to ground answers in the Gemini apps and Vertex AI, and that it has no effect on Google Search. Instead, AI Overviews and AI Mode follow your Googlebot rules, your snippet controls and the Search generative AI control in Search Console.

Does crawl budget matter for a small website?

Rarely. Google’s crawl budget guide (checked October 29, 2026) is written for sites with more than a million pages that change weekly, sites with more than 10,000 pages that change daily, or sites with many URLs stuck in Discovered, currently not indexed. For smaller sites, Google says an up-to-date sitemap and regular Page indexing checks are enough.

Where to Start This Week

Pick three things. Run the four-variant redirect test on a deep URL. Also run the filetype searches for exposed documents. Then open your robots.txt and check each AI user agent line individually. Those three take under an hour combined and they surface the most expensive problems in this list.

After that, work the Critical items in order, then move to High. Do not start at item one and grind forward, because the numbering is a category order, not a priority order.

If you want a second pair of eyes on the results, we run this exact 105-point process for clients. Grab a Free ROI Audit and we will send you the findings with owners and priorities attached.