Is my website AI ready? The fifteen checks

The first mistake is to add an AI widget to your website and call the job finished. A chatbot on your pricing page does not help an assistant understand what you sell, reach the page, or quote it accurately. The cost is a site that looks modern while its important facts remain hard to find.

If you are asking, “is my website AI ready?”, start with the site itself. We ran fifteen readiness checks across 121 audited sites, looking at reach, understanding and parsing.

This is about whether your site is ready for AI assistants, not about adding AI features to it. It measures whether an assistant can get to your pages, work out what they mean and extract useful facts from them. You can run the free AI readiness check, then use the detail below to decide what to fix.

Can AI systems reach your website?

Reach comes first. A clear page a crawler cannot fetch is useless to an assistant, no matter how well it is written.

The reach group covers robots.txt, the sitemap, the homepage, page rendering, and the signals that point crawlers to the preferred page. The sites in this sample generally did well here. The homepage check passed on 119 of 121 sites, and robots.txt passed on 116.

That does not make the group irrelevant. The two homepage failures and five robots.txt failures are small in count, but each can block access to the rest of the work. Rendering also matters because 18 sites failed the check for content being present without JavaScript, while nine more received a warning.

The canonical and Open Graph check passed on 92 sites. It warned on 27 and failed on none. A warning is not the same as a blocked page, but it tells you that the page identity or sharing signals are not clean enough to ignore.

The sitemap check passed on 100 sites and warned on 21. A sitemap is a route map, not a guarantee of inclusion. It still gives crawlers a clearer way to discover the pages you want read.

Does your website explain what you sell?

The understand group asks whether a visitor, crawler or assistant can identify the business and the buying context without decoding brand language.

It includes Organization schema, title and description, a plain business statement, visible pricing or a linked pricing page, and comparison or alternatives pages. These checks expose a larger problem than access alone. Many sites are reachable, but vague.

The plain statement check passed on only 20 of 121 sites. It failed on 43 and warned on 56. Among checks with a full 121-site denominator, this was the weakest result.

A useful statement says what the product is, who it is for, what it costs, and what it is being compared against. It does not need a paragraph of positioning language first.

Pricing was clearer than the wider proposition. The check for a visible price or linked pricing page passed on 90 sites, with warnings on 29 and no failures. That means many sites offer some route to the number, while still leaving room for ambiguity about plans, limits or fit.

Title and description passed on 92 sites, with warnings on 27. Organization schema passed on 72 and failed on 47. Schema helps describe the entity, but it cannot repair a page that never says plainly what the company does.

Comparison and alternatives pages passed on 54 sites. They failed on 21 and warned on 46. This matters when buyers ask an assistant to compare tools, find an alternative or choose between products. A homepage rarely contains enough evidence for that job.

Can AI parse the facts on your pages?

Parsing is the difference between a page being available and a page being usable. The assistant needs clear structure and extractable answers, not merely a successful fetch.

The parse group includes one clear H1, questions answered on the page, llms.txt, Product or SoftwareApplication schema, and whether the site's own pages get cited. The results are mixed.

A clear H1 passed on 101 of 121 sites. Questions answered on the page passed on 56 and warned on 63. llms.txt passed on 66 and warned on 55. Product or SoftwareApplication schema passed on 60 and failed on 59.

The own-citation check had a result for only 21 of the 121 sites. Five passed, 11 failed and five warned. This is a narrow result, and it should not be read as a general probability that your pages will be cited. It says that, in this audit, the site's own pages were rarely the pages appearing as cited sources.

Here is the full set of checks. The checks are unweighted because the data does not provide a different weighting system.

The ledger flags the plain statement, comparison or alternatives pages, and questions answered on the page as the checks that fail or warn most. These are content decisions, not settings you flip once, which probably explains why they lag behind the reach checks.

What should you fix first on your website?

Do not start with the rarest technical setting. Start with the page defects that leave an assistant without a safe answer.

Use this short checklist:

• Write one plain statement that says what the product is, who it is for, what it costs, and what it is being compared with.
• Make the homepage answer a direct request in visible text, without requiring JavaScript to reveal the main facts.
• Add a clear H1 and page title that describe the actual product or service.
• Link to a pricing page, or show the relevant price where a buyer can find it.
• Create comparison and alternatives pages for the choices buyers already make.
• Put common buyer questions and their answers on the relevant page rather than hiding them in navigation or a support system.
• Check robots.txt, the sitemap, canonical tags and Open Graph tags before treating content as the problem.

The plain statement was the rarest pass in the sample. The comparison and question checks also had substantial warning or failure counts, as the table above shows.

Technical fixes still have a place. Product or SoftwareApplication schema had almost the same number of passes and failures. Organization schema also failed on many sites. Add markup where it describes the page accurately, then check that the visible copy says the same thing.

You can use the free AI visibility checker after the page changes. Readiness asks whether the site has the ingredients. Visibility asks whether assistants actually mention or cite it. They are related, and they are not the same result.

What can this audit tell you, and what can it not tell you?

The audit can show which conditions are common across the sample. It can tell you that a plain business statement was uncommon, that homepage reach was common, and that product schema was split between passes and failures. It can give you a sensible order for inspecting your own site.

It cannot tell you that passing a check will make an assistant recommend your business. It cannot identify every crawler, model, query or external source that may affect an answer. It also cannot tell you how your site performs unless you measure your site separately.

We measure samples. We have not tested every website, assistant or market, so we do not know what we have not measured.

The figures here come from 121 audited sites, not from every website, every assistant or every market. The methodology page explains how CiteGraph measures the checks, while the weekly research covers the separate question of who owns AI answers.

The most important caveat is simple: AI readiness is a condition of access and comprehension, not a promise of recommendation or citation.