An AI search audit is a structured review of whether ChatGPT, Gemini, Claude, Perplexity, and other artificial intelligence (AI) assistants can reach a website, read it, describe the company behind it correctly, and cite it in their answers.
The method runs 9 checks in order (crawl access, AI crawler permissions, sitemap quality, structured data, entity signals, answer-first formatting, on-page evidence, third-party citations, and measurement), scores each one 0, 1, or 2, and fixes the lowest scores first.
The audit applies anywhere an assistant composes the answer itself, including Google Search's AI Overviews and AI Mode, OpenAI's ChatGPT search, Anthropic's Claude, Perplexity, and Microsoft Copilot.
They reach your pages through crawlers such as Googlebot, OAI-SearchBot, Claude-SearchBot, and PerplexityBot, and the picture they form of you can draw on your site, schema.org markup, LinkedIn, and review directories.
A site can rank well on Google while a firewall rule or an outdated profile keeps it out of the answers that business-to-business (B2B) buyers read before shortlisting vendors.
In this article you'll learn what an AI search audit is and how it differs from a classic SEO audit, the 9 checks that cover access, how to score each one, and what to fix first so the rest of your work actually shows up in AI-generated answers.
Key Takeaways
An artificial intelligence (AI) search audit checks 9 things in a fixed order, starting with crawler access, because a page an assistant can't fetch can't be cited however strong its content is.
OpenAI documents separate crawlers for model training (GPTBot) and ChatGPT search (OAI-SearchBot) and treats each robots.txt setting as independent, so opting out of training can leave search access intact.
Google sets 2 conditions for a page to be linked as a source in AI Overviews or AI Mode, being indexed and being eligible for a snippet, and says no special schema.org markup or AI text file is needed.
Google ignores the priority and changefreq values in an Extensible Markup Language (XML) sitemap and trusts lastmod only when it is consistently accurate, so sitemap quality comes down to clean uniform resource locators (URLs) and honest dates.
Structured data should be audited for accuracy rather than volume, because Google asks that every marked-up value match the visible text on the page.
A baseline of fixed buyer prompts, recorded before any fix ships, is what lets you show afterward that the audit changed what assistants say about you.
What Is an AI Search Audit?
An AI search audit tests your site and wider web presence against the steps an assistant takes before naming a vendor: fetching pages, reading them, working out who you are, judging trust, and citing a source. Each test produces a score plus a saved piece of evidence, such as a screenshot or an exported report, so anyone can rerun it later.
Know How It Differs From an SEO Audit
A classic search engine optimization (SEO) audit asks whether Googlebot can crawl, index, and rank your pages. Some teams call the newer review a ChatGPT SEO audit or a large language model (LLM) SEO audit, but its scope reaches past your domain.
Audit question | Classic SEO audit | AI readiness audit |
|---|---|---|
Which bots get tested | Googlebot and Bingbot | Googlebot and Bingbot, plus the search crawlers run by OpenAI, Anthropic, and Perplexity |
What gets reviewed | Pages, templates, and rankings | Pages, plus the individual passages an assistant could quote |
Where findings come from | Your own domain | Your domain, LinkedIn, directories, review sites, and publications |
Typical finding | Broken links, slow templates, missing title tags | A search crawler blocked at the firewall, or 3 different company descriptions online |
What it hands back | A prioritized fix list | A scored checklist, an evidence folder, and a prompt baseline |
What Should an AI Readiness Audit Check?
It should check 9 areas that group into access, understanding, trust, and measurement. Work through them in order, because a failure near the top of the list hides every result below it.
Use the 9-Point Checklist
Treat this table as your answer engine optimization (AEO) audit checklist. Copy it into a spreadsheet, add a score column and a notes column, and fill in one row at a time.
# | Check | Test to run | Pass standard | Evidence |
|---|---|---|---|---|
1 | Crawl access | URL Inspection on 10 priority page addresses (URLs), plus bot status codes in content delivery network (CDN) or server logs | All return 200, render main text, and pass the firewall | Screenshots, log extract |
2 | AI crawler permissions | Read robots.txt for OAI-SearchBot, Claude-SearchBot, PerplexityBot, and the * wildcard | No Disallow on pages you want cited | Dated robots.txt copy |
3 | Sitemap quality | Compare the Sitemaps report with your canonical page list | Absolute canonical URLs, no redirects or 404s, honest lastmod | Report export |
4 | Structured data | Test 5 pages in the Rich Results Test or Schema Markup Validator | Valid, and every value matches the page | Validator results |
5 | Entity signals | Compare company facts across About page, LinkedIn, and directories | Matching name, category, products, and buyer | Fact grid |
6 | Answer-first formatting | Copy each section's opening into a blank document | Each opening answers its heading alone | Pass rate per page |
7 | On-page evidence | Highlight every claim on 5 commercial pages | Each claim has a number, date, source, or customer | Marked-up copies |
8 | Third-party citations | Log cited domains for 20 unbranded prompts on 4 assistants | Present on 7 or more of the top 10 cited domains | Citation log |
9 | Measurement | Confirm a fixed prompt set, an AI referral channel, and log access | All 3 exist with a dated baseline | Baseline sheet |
Score Each Check 0, 1, or 2
Give 2 for a clean pass, 1 for a partial pass such as 7 of 10 pages, and 0 for a fail or a check nobody can run yet. Order matters more than the total of 18, so clear every 0 in rows 1 to 3 before touching content, since an assistant that can't fetch a page never sees it.
How Do You Check Whether AI Crawlers Can Reach Your Site?
You test 3 layers: whether pages render for bots, whether robots.txt admits the AI search crawlers, and whether your sitemap points them at the right URLs. AI crawler optimization starts here, because access often breaks through a setting changed for another reason, such as a security rule.
Test Crawl Access Past Robots.txt
A permissive robots.txt means little if your CDN challenges unfamiliar bots or key pages only show text after JavaScript runs. Google's AI features documentation asks owners to allow crawling in robots.txt and at the CDN or host, and to keep important content available as text.
For each priority URL, open the rendered HyperText Markup Language (HTML) in URL Inspection and confirm the main copy is there. Then pull 7 days of server or CDN logs filtered by AI user agents, and use our server log method to tell verified crawlers from impostors.
Audit AI Crawler Permissions One Bot at a Time
OpenAI runs 3 agents: OAI-SearchBot surfaces sites in ChatGPT search, GPTBot collects content for model training, and ChatGPT-User fetches pages during certain user actions. OpenAI's crawler documentation treats each setting as independent and says search changes take roughly 24 hours, so blocking training alone leaves search access as it was.
Anthropic and Perplexity name their own search crawlers, Claude-SearchBot and PerplexityBot. Open yourdomain.com/robots.txt, write down every User-agent group, and flag any Disallow that reaches a page you want cited, including rules under the * wildcard that apply to every bot.
Grade Sitemap Quality Against Google's Rules
Google's sitemap guidance caps each sitemap at 50,000 URLs or 50MB uncompressed, asks for fully qualified absolute URLs, and uses lastmod only when it is consistently and verifiably accurate. Google also ignores priority and changefreq, so tuning those values buys nothing.
Compare the Sitemaps report export with your canonical page list and flag every redirect, 404, noindexed URL, and lastmod date that resets on each deploy. Confirm robots.txt carries a Sitemap line, and use our infrastructure checklist for canonical tags and internal linking.
Does Structured Data Matter for AI Search?
Structured data labels facts like your company name, authors, and publish dates for machines, but Google states that AI Overviews and AI Mode need no special schema.org markup. When you audit structured data for AI search, check accuracy rather than volume.
Match Every Marked-Up Value to the Page
Google's AI features guidance asks that structured data match the visible text on the page. Run 5 priority pages through Google's Rich Results Test or schema.org's Schema Markup Validator, then compare each value, such as price, rating, author, date, and address, with what a visitor actually sees.
Mismatches tend to appear after a price change or a rebrand, when the copy changed and the template didn't. Score the check 0 if any value contradicts the page.
How Do You Audit Entity Signals?
You list your core company facts, then check every place they appear, including assistants' answers about you. The goal is one consistent description that assistants can match to the right buyer questions.
Build a Fact Grid Across Sources
Make a grid with 6 rows (name, category, main products, target buyer, headquarters, and pricing model) and 1 column per source: your About page, LinkedIn, Crunchbase, your main review directory, and your press boilerplate. Mark every cell that disagrees with the About page.
Score the check 2 if no cell disagrees, 1 for small wording drift, and 0 if any source names an old product, category, or buyer. Fix your own pages first, then send each profile owner the corrected wording.
Ask Each Assistant to Describe You
Ask ChatGPT, Gemini, Claude, and Perplexity what your company does, who it sells to, and what it costs, with sources. Paste each answer into a new grid column to see where every assistant's version departs from yours and which page it cited.
How Do You Audit Content for Answer-First Formatting and Evidence?
Pick your 10 most commercial pages and run 2 tests on each: whether its passages stand alone and whether its claims carry proof. Both read the page the way an assistant quoting one passage would.
Lift Each Opening Out of Context
Paste the opening 2 sentences of each section into a blank document, leaving the heading behind. The passage fails if it leans on words like "this" or "as mentioned" for meaning, or never answers the heading.
Divide passing openings by total openings and set a pass mark, such as 8 in 10. Rewriting openings costs less than new pages, so this fix often ships first.
Score the Evidence Behind Each Claim
On your 5 main commercial pages, highlight every claim about results, speed, savings, or customer outcomes. A claim passes when it carries a number, a date, a named source, or a named customer, and a visible author and publish date add context to the page as a whole.
Score each page by the share of claims that pass, and list each unsupported one with the proof it needs. Adjective-heavy pages give an assistant nothing specific to repeat, so this list often runs longest.
How Do You Audit Third-Party Citations?
Run your category prompts on each assistant, record every domain the answers cite, and compare that list with where your brand appears. The gap between the 2 lists becomes your off-site fix list.
Log Cited Domains by Assistant
Write 20 unbranded prompts from sales calls and demo forms, such as requests for a vendor shortlist in your category. Run each in a fresh session on ChatGPT, Gemini, Claude, and Perplexity, log every cited domain beside its assistant and prompt, and see our analysis of most-cited domains for how those lists shift by industry.
Rank the Gaps by Overlap
Sort the log by how many assistants cite each domain. Domains cited by 2 or more assistants that leave you out go to the top of the fix list, followed by domains listing you with outdated facts.
Score the check 2 if you appear on 7 or more of the 10 most-cited domains, 1 for 3 to 6, and 0 for fewer than 3. A new listing counts only once an assistant starts citing it, so rerun the prompts after each outreach push.
How Do You Measure Whether the Audit Worked?
Record a baseline before you fix anything, then rerun the same prompts and checks on a schedule. Without that ninth check, the other 8 can't show progress.
Lock a Baseline Prompt Set
Save the 20 prompts from the citation check, add a few branded ones, and record for each assistant whether you were mentioned, cited, or recommended. Keep the list unchanged between runs and use our guide to buyer prompt research to choose them.
Read the Reports You Already Have
Search Console counts AI Overviews and AI Mode traffic in the Performance report under the Web search type, and its generative AI performance report shows impressions by page, country, device, and date. ChatGPT tags referral links with utm_source=chatgpt.com, so group AI referrals into one Google Analytics 4 (GA4) channel and compare it monthly.
Book a Free AI Visibility Audit With Windgrove
Windgrove runs this review for B2B software as a service (SaaS), fintech, infrastructure, and technology companies, starting with a free diagnosis. When the findings call for fixes, our managed programs carry them through.
Start With the Free Diagnosis
The free AI Visibility Audit maps your brand's presence in AI search, the competitors ahead of you, the buyer queries you don't yet appear for, and the fixes to make first. Request a free audit, then use the 9-point checklist above to work through the findings.
Hand Us the Fixes
Every Windgrove retainer includes technical AEO and SEO work, content, authority building, and AI visibility tracking, and each starts on a 90-day initial term. Claim Your Ground, the entry program, is $4,200 per month for 50 tracked buyer prompts, with a 90-day guarantee of a +20% minimum increase in unbranded AI visibility over the day-one baseline. If the target is missed, we work for free until we hit it. You can compare all 3 programs before booking a call.
See What Fixing the Foundation Did for Opal
When fintech client Opal Spend began working with us, it had no blog, no sitemap, and only 4 indexed pages. We fixed the sitemap, robots.txt, metadata, and indexation before any content shipped. Tracked AI brand mentions across its core prompts then rose from 3 to 39, and by the 90-day mark AI search was bringing in 10+ inbound demos each month.
Frequently Asked Questions (FAQs)
Can I Run an AI Readiness Audit Myself?
Yes. Every check here runs on free tools: Google Search Console, a browser, the Rich Results Test or Schema Markup Validator, server logs, a spreadsheet, and the assistants themselves. The harder part is fixing what the audit finds, especially firewall rules, markup, and third-party profiles that need a developer or outreach.
Should I Block GPTBot to Protect My Content?
That depends on whether you want your content used for model training. OpenAI documents GPTBot as its training crawler and OAI-SearchBot as the crawler behind ChatGPT search, each set independently. Blocking GPTBot alone leaves ChatGPT search access unchanged, while blocking OAI-SearchBot keeps your pages out of those answers.
Do I Need Schema Markup to Appear in AI Overviews?
No special markup is required. Google's documentation sets 2 conditions for a page to be linked as a source in AI Overviews or AI Mode: being indexed and being eligible for a snippet. It adds that no special schema.org structured data is needed, though accurate markup still describes your company clearly.
How Often Should I Repeat the Audit?
Run the full 9-point audit once a quarter and after any migration, redesign, content management system (CMS) change, or new firewall setting, since those moments break access rules unnoticed. Rerun the prompt baseline monthly with identical prompts, so you can tell whether visibility moved because of your fixes.
How Is an AI Readiness Audit Different From an SEO Audit?
A search engine optimization (SEO) audit checks whether search engines can crawl, index, and rank your pages. An AI readiness audit adds the AI search crawlers, passage-level formatting, entity consistency across third-party profiles, and the domains assistants cite, then tracks mentions and citations across fixed prompts.
What Does Windgrove's Free AI Visibility Audit Cover?
Windgrove's free AI Visibility Audit reviews how visible your brand is in AI search, which competitors sit ahead of you, the buyer queries you don't yet show up for, and the opportunities to tackle first. It costs nothing, and paid programs that handle the fixes begin at $4,200 monthly on a 90-day initial term.