The AEO Checklist: 8 Things AI Engines Check Before They Cite You

The short version:

• Before an AI engine can name your business, it has to clear three stages in order: reach your page, read your page, and trust what it found. Most sites fail at stage one or two and never get judged on quality at all.
• The eight checks below are the ones that decide it, listed in the order they actually break. Fixing them out of order wastes time, because schema on a page no crawler can fetch does nothing.
• The two most commonly missed are ChatGPT-specific: allowing OAI-SearchBot (its live search crawler, which is not GPTBot) and being present in Bing's index, which still feeds ChatGPT's search.
• In our audit of 50 funded Toronto startups the median score was 10.75 out of 20. Most of the lost points were on checks 5 through 8, the deliberate ones.
• You can run all eight against your own site in about two minutes with our free AEO checker.

We built this list backwards, from the failures. Every item here is something we have watched cost a real business its place in an AI answer, and every one of them is fixable without a rebuild.

What do AI engines check before citing a site?

Three things in sequence. Can they fetch the page (crawler permission, CDN rules, index presence). Can they read it (raw HTML content, not JavaScript-rendered). Can they trust and lift it (a direct answer, machine-readable schema, a consistent identity, reasonably current information). A failure at any earlier stage makes the later ones irrelevant.

That sequencing is the part most checklists get wrong. They lead with content advice, which is the fun part, when the business in front of them is serving a blank page to every crawler on the internet. Work the list in order.

Stage one: can they reach you

Check 1. Do you allow the AI crawlers, including OAI-SearchBot?

In robots.txt, allow OAI-SearchBot (the crawler behind ChatGPT's live search), GPTBot (OpenAI's training crawler), ClaudeBot, PerplexityBot, Google-Extended, and CCBot. The distinction that trips people up is that OAI-SearchBot and GPTBot are different bots with different jobs. Blocking GPTBot to keep your content out of training does not block ChatGPT search, and blocking OAI-SearchBot removes you from ChatGPT's answers entirely.

We see this exact mistake in both directions. Some businesses block everything OpenAI-shaped on principle and then wonder why ChatGPT never names them. Others allow GPTBot only, assume they are covered, and are still invisible in search. If you want the citations but not the training use, allow OAI-SearchBot and disallow GPTBot. That is the correct configuration, and almost nobody has it.

In our audit the good news was that roughly 90% of Toronto startups did allow AI crawlers. This check is the one most sites already pass.

Check 2. Is your CDN quietly blocking them anyway?

Yes, and this is the failure that hides best. Cloudflare and similar providers ship bot-management rules that can challenge or block AI crawlers at the network layer, before robots.txt is ever consulted. Your robots.txt says come in, your CDN says no, and you never see an error because nothing broke for human visitors.

The reason this one is worth its own check is that the person who wrote the robots.txt is usually not the person who configured the CDN. We have found sites with a beautifully permissive robots.txt sitting behind a bot rule that returned a challenge page to every AI crawler that tried. The fix is a few minutes in the dashboard once you know to look. Finding out is the hard part, which is why our checker fetches your page the way a crawler does rather than trusting your robots.txt at face value.

Check 3. Are you in Bing's index?

Yes. ChatGPT Search launched on Bing's index and still runs on Bing's crawling infrastructure, now as a re-ranked hybrid with OpenAI's own OAI-SearchBot index and undisclosed partners. Bing's measured share of ChatGPT's sourcing has reportedly fallen over the past year, but a site Bing cannot find is a site ChatGPT often cannot find either. Registering in Bing Webmaster Tools and submitting URLs through IndexNow is free and takes about ten minutes.

This is the single most skipped step in every ChatGPT visibility guide we have read, and we think that is because Bing has been unfashionable for fifteen years. Marketers optimise for Google out of habit and never open a Bing account. The irony is that the least glamorous property on the internet is now a partial gatekeeper to the most talked-about one. Ten minutes, no cost, and you can import your existing Google Search Console property directly rather than starting from scratch.

Stage two: can they read you

Check 4. Does your page render without JavaScript?

It has to. Most AI crawlers do not execute JavaScript, so if your homepage is a client-rendered app that paints content after load, the crawler sees an empty shell. You can test this yourself in thirty seconds: disable JavaScript in your browser settings, reload your homepage, and read what is left. If the page is blank or shows only a loading spinner, nothing else on this list matters yet.

This is the check with the highest cost of failure and the widest range of fix effort. On a WordPress or Webflow site it is usually a non-issue. On a single-page React app with no server rendering it is a genuine engineering project, and we would rather tell you that honestly than sell you schema work that will never be seen. In our audit 93% of companies passed this one, which is better than the industry reputation suggests, but the 7% that failed were failing completely rather than partially.

Stage three: can they trust and lift you

Check 5. Do you have an answer-first passage?

A short block, roughly 40 to 60 words, that directly answers the question a customer would ask, placed immediately under the heading that asks it. Not a preamble, not brand positioning, just the answer stated plainly enough that a model can lift it verbatim and be confident it is correct. Every key page should have one.

The test we use is blunt: could an AI quote your first paragraph as the answer to a real customer question without editing it? Most homepages fail because they open with a slogan. "Transforming how modern teams work" tells a machine nothing about what you do or who you serve. "We are a Scarborough family dental clinic accepting new patients" is quotable. In our audit only 30% of companies passed a plain version of this test.

The related discipline is structuring the whole page around the questions people actually ask rather than the topics you want to talk about. That is the same reasoning behind writing for query fan-out, where an AI breaks one question into several and looks for a source that answers each cleanly.

Check 6. Do you have Organization and FAQPage schema?

At minimum, Organization schema (your name, location, services, contact details, and the links that confirm your identity) and FAQPage schema on pages that answer real customer questions. Schema is how you state facts about yourself in a form no machine has to infer. Everything else on your page is prose a model must interpret; schema is a fact it can read directly.

This is where Toronto is weakest by a wide margin. Only 5% of the funded startups we audited published FAQPage schema, and 52% were missing basic Organization schema entirely. Those are companies with funding, engineering teams, and marketing budgets. The gap is not capability, it is that nobody told them it mattered.

We wrote separately about the flip side of this, that a third of the same companies had shipped an llms.txt file that no major engine commits to reading. The city is triple the global rate on the file that does nothing and one in twenty on the schema that works.

Check 7. Is your identity consistent across the web?

Because engines cross-reference. When your website, Google Business Profile, LinkedIn, and directory listings all describe you the same way with the same name, address, and service description, a model can resolve you to one confident entity. When they disagree, it hedges, and hedging usually means naming a competitor it is surer about.

The everyday version of this failure is mundane. A suite number on the website that is missing from the Business Profile, a service listed on LinkedIn that appears nowhere else, an old business name still live in three directories. None of it looks like a marketing problem. All of it makes you harder to be certain about, and certainty is what gets you named.

Check 8. Is your content current enough to be cited?

It helps. Engines favour content that is clearly dated and recently updated, particularly for anything with a year, a price, or a regulation attached. A page with a visible publish or update date and current information reads as safer to quote than an undated page that might be from 2019.

The practical version is not a content treadmill. Put real dates on your pages, review the handful that carry time-sensitive facts a couple of times a year, and update them when the facts change. That is enough. We would rather you spend the effort on checks 5 and 6, which are worth more.

How do I check all eight at once?

Yes. Our free AEO checker runs these checks against your site automatically. It fetches your page the way an AI crawler does, tests whether your content survives without JavaScript, verifies OAI-SearchBot access and whether you are present in Bing's index, looks for Organization and FAQPage schema and an answer-first structure, and returns the specific gaps rather than a score you cannot act on.

Once you have the list, work it in the order above: reach, then read, then trust. And if you want to know whether the work moved anything, our monthly Toronto AI Citation Tracker shows which engines are naming local businesses and how that shifts month to month. For the strategic context behind all of this, AEO vs SEO covers what genuinely changes when your buyers ask an AI instead of Google.

The honest summary

Six of these eight checks are an afternoon of work. One (the JavaScript render) is either trivial or a real project depending on how your site was built, and one (entity consistency) is tedious rather than hard. That is the entire barrier to being eligible for AI answers right now, and most of your competitors have not cleared it. That will not be true for long.

Sources

1. Stratezik Toronto Startup Website Audit 2026: 50 funded Toronto and GTA startups on a machine-verified 20-point AEO test (90% allow AI crawlers, 93% render without JavaScript, 30% pass answer-first, 5% FAQPage schema, 52% missing Organization schema, median 10.75 of 20). Dataset available on request.
2. ChatGPT Search sourcing as a re-ranked hybrid of OpenAI's OAI-SearchBot index, Bing's crawling infrastructure, and undisclosed partners, with OAI-SearchBot powering live search while GPTBot is training-only: Yoast, "What is ChatGPT Search"; AI+Automation, "Does ChatGPT Use Bing or Google".
3. llms.txt adoption and citation-prediction findings (SE Ranking study across nearly 300,000 domains, as analysed by OrganiKPI): organikpi.com/blog/distribution/llms-txt-adoption-impact.
4. Stratezik Toronto AI Citation Tracker, July 2026. Available at stratezik.com/blog/toronto-ai-citation-tracker-july-2026.

• $3,000 Growth CreditFree assessment · credit applied to onboarding. See if you qualify →

Want the optimization playbook, not just the platform overview? Get our free ChatGPT Ads cheat sheet — context hints, bid-floor tests, industry readiness, and the measurement stack from practitioners running real budgets.

Quick answers