The AI Search Audit: A 10-Step Checklist for Your Site

What are the ten checks, and in what order?

Run these ten checks in order, because each one assumes the one before it passed. There is no point checking whether your content answers a question if the crawler that would read it is blocked at the gate, and no point auditing brand accuracy before confirming the engines can reach your pages at all. The list below moves from access, to content, to competitive position, to a baseline you can compare against next quarter.

1. Crawler access

What to check: Open yoursite.com/robots.txt and look for Disallow rules against the crawlers that feed AI answers: OpenAI's GPTBot and OAI-SearchBot, Anthropic's ClaudeBot and Claude-SearchBot, plus whichever of Google's crawlers you use for Search. (ChatGPT-User and Anthropic's Claude-User are triggered by a live user request rather than automated crawling, so a robots.txt block on them may not even apply — check them anyway.)

Pass/fail: Pass if none of the above is disallowed on the pages you want cited. Fail if any one is blocked anywhere on the site you care about — including a rule that was meant to be temporary.

This is the single most common self-inflicted gap; see should you block AI crawlers like GPTBot for the tradeoffs before you change anything.

Example row (filled in): Audit of a fictional site, run 9 October 2026. Checked lumenanalytics.example/robots.txt. Found:

User-agent: GPTBot
Disallow: /blog/

Result: FAIL. OpenAI's training crawler is blocked from the blog, which is exactly where buying-intent content lives. Fix: remove the Disallow line for GPTBot and OAI-SearchBot unless deliberately opting the blog out of model training, in which case block only GPTBot and leave OAI-SearchBot (the ChatGPT search/citation crawler) untouched.

2. Rendering

What to check: Fetch the raw HTML of a key page — curl it, or use your browser's "view source," not the rendered DOM in dev tools — and see whether your main heading and body copy appear in that raw response, or only after JavaScript runs.

Pass/fail: Pass if the primary content is present in the initial server response. Fail if the raw HTML is a near-empty shell and the content only appears client-side.

3. Entities

What to check: Is your brand name, product name and one-line category description consistent across your homepage, your About page, and the handful of third-party sources an engine is likely to pull from (Wikipedia or Wikidata if you have an entry, LinkedIn, Crunchbase, G2)? Check for Organization schema too.

Pass/fail: Pass if the name and description match across at least three or four sources. Fail if your homepage, your LinkedIn and your directory listings each describe you differently, with no single consistent version anywhere.

4. Answer-ready content

What to check: For each of your three to five core pages, can you point to one sentence near the top that directly answers the question a buyer would type — "What is X," "How much does X cost," "Who is X for"?

Pass/fail: Pass if a standalone sentence answers the question within the first two paragraphs. Fail if the answer is implied, scattered across the page, or buried under scene-setting copy.

5. Schema

What to check: Run your most important pages through a structured-data validator (such as Google's Rich Results Test) for Organization, Product, FAQPage or Article markup.

Pass/fail: Pass if valid schema is present with no errors on the pages that matter most. Fail if there's no schema, or schema that throws errors. Note that Google's own guidance says structured data isn't required for its generative AI features specifically — this check is about general findability and rich-result eligibility, not a generative-AI requirement.

6. Buying-intent prompts and cited sources

What to check: Ask five to ten buying-intent prompts in your category — without your brand name — on ChatGPT, Gemini, Claude and Perplexity, in fresh, logged-out sessions. Log which sources get cited for each.

Pass/fail: Pass if at least one of your own pages is cited as a source in at least one engine for at least one prompt. Fail if only competitors' pages ever show up. See how to measure AI visibility for how to structure this logging so one lucky run doesn't look like a trend.

7. Brand accuracy

What to check: Ask each engine directly: "What is [brand]?" and "What does [brand] cost / do / serve?" Compare the answers to your actual facts.

Pass/fail: Pass if there's no material factual error — wrong price, a dead feature, the wrong audience. Fail if at least one answer states something false or outdated about you. Full method here: how to check what ChatGPT says about my brand.

8. Competitor gaps

What to check: Using the same prompts from step 6, note every source cited for a competitor that never appears for you — comparison pages, review sites, documentation, forum threads.

Pass/fail: Pass if you share most of the same cited sources as your competitors. Fail if competitors are consistently backed by a type of source — reviews, comparisons, docs — that your brand has none of anywhere.

9. Third-party presence

What to check: Search your brand name plus "review," "alternative" and "vs," and see whether you appear on the directories, comparison sites and forums that typically surface as citations in your category — G2, Capterra, Reddit, trade directories.

Pass/fail: Pass if you're present and reasonably current on at least two or three of the sources your category's engines tend to cite. Fail if you're absent from all of them, or present with stale or wrong information.

10. AI Overviews baseline

What to check: Search three to five of your highest-intent Google queries, logged out and in an incognito window. Record, with a screenshot and the date, whether an AI Overview appears and whether you're named or linked in it. Google's own guidance points to the Search Console "generative AI performance report" as a way to monitor this over time.

Pass/fail: Pass simply means a dated baseline now exists, whatever it shows. Fail means the step was skipped — the point of this one is having something to compare against next quarter, not a verdict today. For the mechanics of Google's generative features specifically, see the Google AI search optimization guide.

How to score it, and what to fix first

Count your passes out of ten.

• 8–10: A solid technical foundation. Spend your next afternoon on steps 6–9 — the competitive and prompt-level work — rather than re-checking access.
• 5–7: Real gaps worth prioritizing before any content or outreach push.
• Below 5: Stop and fix steps 1 and 2 before anything else. A perfect answer on a page a crawler can't reach is invisible regardless.

Fix order, independent of score: crawler access and rendering (1–2) always come first, because every other check assumes a bot can reach and parse the page. Entities and schema (3, 5) come next — they're cheap, one-time fixes. Answer-ready content (4) follows. Steps 6–9 require ongoing prompt testing and real-world presence-building, so treat them as the recurring half of the audit rather than a one-time fix. Step 10 is a baseline you can log at any point, but keep it dated so it's comparable later.

How to repeat this audit

Re-run the full ten steps after any robots.txt change, CMS or template migration, or major rebrand — any of those can silently flip a pass to a fail. Otherwise, a quarterly cadence is reasonable: AI answers vary run to run even when nothing on your site changed, so a single pass or fail on steps 6–10 is a snapshot, not a permanent score. Keep each run's dated log rather than overwriting the last one, so you can see whether a gap is closing or just noisy.

FAQ

How long does this audit actually take? Most of it fits into a single afternoon: robots.txt and view-source checks take minutes each, the schema validator run is quick, and the prompt-based steps (6–8) are the only part that takes real time, since each needs several prompts across more than one engine.

Do I need any paid tools to run it? No. A robots.txt file, your browser's view-source, a free structured-data validator, and chat interfaces for the major engines cover all ten steps.

Which check should I fix first if I only have time for one? Crawler access. If a bot is blocked, nothing else on this list matters, because the content behind that block is invisible regardless of how good it is.

Is this the same as ContextOwl's own technical audit? No. This is a general, free checklist anyone can run by hand in an afternoon. ContextOwl's full report runs a separate, deeper 11-check technical audit alongside prompt-based testing, delivered as a paid one-time report — the two lists aren't interchangeable.

What if an engine gives a different answer the next time I run the same prompt? That's expected, not a bug in the audit. Treat steps 6–8 as a snapshot and log the date; a single run passing or failing is a data point, not a verdict, which is why repeating the audit matters more than any one pass.

What to do next

1. Block an afternoon and run all ten checks against your own site, logging each as pass or fail with today's date.
2. Fix crawler access and rendering first — every other item on the list depends on a bot being able to reach and parse the page.
3. Re-run the full audit next quarter, or immediately after any robots.txt, CMS or template change, and compare against today's log.

This checklist is general guidance, not a report on your specific site; for a quick read on whether AI crawlers can reach yours, try the AI Bot Access Checker. Selling into India specifically? See the D2C India AI search visibility checklist for the market-specific version of this list.