TL;DR: Users click a traditional result on 8% of visits when an AI summary appears, versus 15% without one (Pew Research Center, 68,879 real searches, March 2025). The response is not panic; it is an audit. These are the 9 checks we run to determine whether a site can earn AI citations, in the order we run them, each with its fix.
An AI SEO audit answers one question: when an AI system retrieves sources for a query you should own, is your site selectable, and if not, which specific gate is failing?
The stakes are measured, not hypothetical. Alongside the Pew click data above, Ahrefs found AI Overviews correlated with a 34.5% lower clickthrough rate for top-ranking pages across 300,000 keywords (April 2025). The other side of the ledger: citation has come unchained from ranking, with 36.7% of AI Overview citations now going to pages outside the top 100 (Ahrefs, March 2026). Sites that fail this audit lose clicks with no citation offset. Sites that pass it can win visibility they never earned in classic rankings.
Prerequisites: access to your robots.txt, Google Search Console, Bing Webmaster Tools, and GA4. Nothing here requires a developer for diagnosis; a few fixes do.
Check 1: Are your pages indexed and snippet-eligible?
The entire technical bar for Google's AI features, per Google's own documentation: "a page must be indexed and eligible to be shown in Google Search with a snippet. There are no additional technical requirements."
Check: run your money pages through Search Console's URL Inspection. Then check your page source and CMS settings for "nosnippet," "data-nosnippet," or "max-snippet" directives you don't remember adding; SEO plugins sometimes set them.
If it fails: an unindexed page fails every other check automatically, so fix crawlability first. A snippet restriction is a one-line removal, but per Google the same controls govern AI features and classic snippets, so removing it re-opens both surfaces at once.
Check 2: Are AI crawlers reaching your site at all?
Four robots.txt tokens decide most of your AI visibility, and they do different jobs. Per each company's official crawler documentation:
- OAI-SearchBot (OpenAI): controls ChatGPT search. "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers."
- GPTBot (OpenAI): controls model training, not search. The two are independent; you can allow search visibility while refusing training.
- PerplexityBot (Perplexity): controls Perplexity search surfacing, and is "not used to crawl content for AI foundation models."
- Google-Extended (Google): controls Gemini training and grounding, and "does not impact a site's inclusion in Google Search nor is it used as a ranking signal." Blocking it does not opt you out of AI Overviews; snippet controls do that (Check 1).
Check: read your robots.txt, then check your CDN and security plugin for bot-blocking rules. Cloudflare-class "block AI bots" toggles are the most common silent killer we find; the site owner never chose to disappear from ChatGPT search.
If it fails: unblock the search crawlers (OAI-SearchBot, PerplexityBot), decide the training question (GPTBot, Google-Extended) as a separate business decision, and verify real crawler traffic against the published IP lists before whitelisting user-agent strings anyone can spoof.
Check 3: Does each page answer its question in the first 100 words?
AI systems select passages, not pages, and 90% of top-cited passages state the answer within the first 100 words (LLMClicks data, compiled by ZipTie.dev, 2025–2026).
Check: open your five most important pages and read only the first 100 words of each. Could that fragment be lifted into an AI answer and stand alone? Or does it warm up with context and promise the answer later?
If it fails: restructure the opening, answer first, support after. If a page cannot state its answer in 100 words, it is usually answering too many questions at once; split it.
Check 4: Do your headings match the questions buyers actually ask?
Google's AI features use query fan-out: the system decomposes a query into sub-questions and retrieves the best answer to each. Your headings are the map it reads.
Check: pull your Search Console query report and the People Also Ask boxes for your target terms. Compare them to your H2s. If your headings are labels ("Our Process," "Key Benefits") instead of questions a retrieval system can match, you are invisible at the passage level.
If it fails: rewrite headings as the questions themselves and give each sub-question its own self-contained section, the answer-capsule discipline this cluster is built on.
Check 5: Is schema markup present where it actually earns extraction?
Honest framing first, because vendors muddy this: per Google, "there's also no special schema.org structured data that you need to add" for AI Overviews. Schema is not an eligibility gate (the full breakdown: schema markup for AI search). What it changes is extraction: schema-marked pages took top-3 Perplexity citation slots at 47% versus 28% without (Onely data, compiled by ZipTie.dev, 2025–2026).
Check: validate your money pages in Google's Rich Results Test. FAQPage on pages with genuine Q&A content and Article/BlogPosting with author fields on posts are the two that matter.
If it fails: add JSON-LD via your CMS. Skip anyone selling proprietary "AI-readiness" markup; if it isn't on schema.org, no platform has documented reading it.
Check 6: Does every stat in your content carry a source and a date?
Both retrieval systems and human buyers treat an unsourced number as noise. A passage whose facts arrive pre-verified, with a named source and a date, is cheaper for an AI system to select and safer for it to quote.
Check: scan your key pages for numbers. Every one should answer "says who, and when?" inline.
If it fails: source it or cut it. Anything older than 18 months gets refreshed or labeled as legacy data. This is also the E-E-A-T substance Google's raters and AI systems both screen for.
Check 7: Is your refresh cadence matched to how fast your topics age?
About 50% of Perplexity's citations point to content published in the current year (Seer Interactive, 5,000+ URL study, June 2025). Freshness is a live selection signal on every retrieval platform.
Check: list your ten most valuable pages with their last substantive update. Anything with rates, laws, pricing, or product comparisons that hasn't been touched in over a year is decaying inventory.
If it fails: build a refresh calendar by decay speed: quarterly for time-sensitive money pages, annually for evergreen definitions. Update the substance and the dateModified field, not the year in the title.
Check 8: Do you have content no AI could have written from search results?
An AI system assembling an answer from interchangeable summaries can cite any of them. It cites the irreplaceable one when it exists: first-party numbers, documented client patterns, before-and-after outcomes.
Check: for each money page, ask what sentence on it could only have been written by someone who does the work. If the answer is none, the page is a paraphrase competing with a million paraphrases.
If it fails: this is the strategy problem to solve before writing anything new. One post built on our own operator data generated 932+ AI citations from a zero-authority domain; the method is the substance, not the formatting.
Check 9: Are you measuring AI visibility at all?
You cannot manage what you cannot see, and most sites cannot see any of this.
Check: three instruments, all free. Bing Webmaster Tools' AI Performance dashboard (launched February 10, 2026) shows per-URL citation counts, the only direct citation console any platform offers. GA4 referral segments for perplexity.ai, chatgpt.com, and copilot.microsoft.com show AI-driven visits. In Search Console, impressions climbing while average position stays flat is the fingerprint of AI-surface visibility, since Google folds AI Overview traffic into the Web search type without a breakdown.
If it fails: set up all three this week, before making content changes, so you have a baseline. On our own blog these instruments are how we watched citations climb from zero to a 171-per-day peak while Google organic clicks were still near zero; without them, that entire result would have been invisible.
What does passing look like?
The success criterion for the whole audit is one sentence: for each query you should own, your page offers a self-contained, sourced, current passage that an AI system can lift without editing, on a site its crawlers can reach, with instruments in place to watch it happen.
On this blog, that standard took a new domain with minimal backlinks to 932+ Bing AI citations, settling at 56–111 per day, with cited pages diversifying from 2–4 per day to 9–15 as newer posts earned slots. The audit above is the same one we run for clients; the checks are identical, the operator data changes.
Frequently Asked Questions
How long does an AI SEO audit take to run?
Diagnosis is a half-day with this checklist: robots.txt and snippet checks are minutes, structure and sourcing reviews take a couple of hours across your top pages, and instrument setup is under an hour. Remediation is the variable: heading and opening-paragraph rewrites are days of work, while building genuine operator content (Check 8) is an ongoing practice, not a task.
How often should the audit be repeated?
Quarterly for the technical checks, because CDN rules, plugin updates, and new AI crawlers change the access picture without notice. The content checks (structure, sourcing, freshness) belong in your refresh calendar rather than a standalone re-audit. Re-run everything immediately after replatforming or changing CDN or security vendors; that is when silent crawler blocks appear.
Do I need separate audits for Google, ChatGPT, and Perplexity?
One structural audit covers all three, because the selection signals overlap heavily. What differs per platform is access (Check 2's per-crawler tokens) and measurement (Check 9's per-platform instruments). Budget one audit, then track each platform separately, since being cited on one predicts almost nothing about the others.
Is an llms.txt file worth adding?
It won't hurt, but nothing audited here reads it. Google states plainly: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." OpenAI's and Perplexity's crawler documentation likewise defines behavior via robots.txt tokens, not llms.txt. Treat it as optional garnish, and be skeptical of any audit that leads with it.
What's the single highest-impact fix on this list?
For sites already publishing decent content: Check 3, rewriting openings to answer first, because it upgrades existing pages without new production. For sites with a blocking problem in Check 1 or Check 2, nothing else matters until access is fixed; a blocked crawler makes every content improvement invisible.
Falling informational clicks are the part of AI search you don't control. Whether your site is selectable when the AI systems come looking is the part you do. And if the audit shows you're invisible right now, the seven mistakes that cause it are the usual culprits, in order.
We run this exact audit, with the citation data to show what passing produces: 932+ AI citations on a domain that had no authority when it started.
Get the ClickWerxs AI SEO audit →
Kaleb Dickhaut — Founder, ClickWerxs. Kaleb built ClickWerxs from the ground up — from payment processing ISO to the Command Center platform to the AI SEO methodology the blog runs on. He has onboarded hundreds of small businesses onto payment and CRM systems. linkedin.com/in/kaleb-dickhaut
Sources
- StatCounter Global Stats, search engine market share worldwide — Google 91.25% as of June 2026. gs.statcounter.com
- ClickWerxs blog, first-party AI citation data, May 2026 — one post generated 932+ Bing AI citations across query variants from a new domain; daily citations rose from 0 to a peak of 171 on 7 May 2026, settling at 56–111 per day, with Google organic clicks near zero over the same period. Reported as a past result, not a promise of performance. Operator data.
ClickWerxs sells SEO and AI visibility services and earns revenue from those engagements. First-party figures are past results for this blog and are not a promise of future performance. This is operator opinion, not professional advice.