CoreCited
Log inStart free
HomeFree toolsFeaturesGuidesBlogPricing

How to optimise for AI search, as an actual checklist

Every item mapped to one of the 31 audit checks, with why it matters and how to verify it yourself.

10 min read

To optimise a page for AI search, make sure AI crawlers can reach it, that its content is in the HTML, that it opens with a direct answer, and that its structure, dates and brand name are clear to software. Below are 31 specific checks, each with why it matters and how to verify it yourself.

These are the same 31 checks the CoreCited site audit runs, with the same pass thresholds. You do not need our tool to use them. Every one can be checked with a browser, View Source and PageSpeed Insights.

One honest note first. Google says there are no special requirements or markup for its AI features. Nothing here is a secret switch. These checks remove the common reasons a page gets skipped: an engine cannot fetch it, cannot parse it, or cannot find a clean passage to quote.

Can AI reach the page?

  1. AI crawlers are not blocked in robots.txt. An engine cannot cite what its crawler is not allowed to fetch. How to check: Open yoursite.com/robots.txt and look for Disallow: / under GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot or User-agent: *.
  2. The page is indexable. A noindex tag tells every engine to ignore the page entirely. How to check: View the page source and search for noindex in the robots meta tag.
  3. The content is in the HTML, not only added by JavaScript. Many AI crawlers do not run JavaScript. If the text only appears after scripts run, they see an empty page. We pass a page with at least 800 characters of server-rendered text. How to check: View source (not Inspect) and check your main text is actually there.
  4. A canonical URL is declared. It tells engines which URL is the real one, so signals are not split across duplicates. How to check: Search the source for rel="canonical".
  5. The URL returns a clean 200. Every redirect hop loses a little, and some crawlers stop following. How to check: Open the exact URL you link to and confirm it does not redirect.

Can AI understand what the page is?

  1. The page has JSON-LD structured data. It states what the page is (an article, a product, a FAQ) in a format built for software. How to check: Search the source for application/ld+json.
  2. The JSON-LD parses and declares a type. A malformed block is ignored completely. How to check: Paste the page into Google's Rich Results Test or the Schema.org validator.
  3. Organization or WebSite schema is present. It connects the page to your brand as a named entity, with your name, URL and logo. How to check: Look for "@type": "Organization" or "WebSite" in the JSON-LD.
  4. FAQ schema is present where there are questions. Question-and-answer pairs are the shape answer engines extract most easily. How to check: Look for "@type": "FAQPage".
  5. Schema blocks have a name and description. A typed block with no properties tells an engine nothing. How to check: Check each JSON-LD block has name and description filled in.
  6. Breadcrumb schema is present (except on the homepage). It shows where the page sits in your site. How to check: Look for "@type": "BreadcrumbList".

Can AI answer from the page?

  1. There is exactly one H1. None leaves the topic unclear. Several make it contradictory. How to check: Search the source for <h1 and count.
  2. The page has at least two H2 headings. Engines lift answers section by section. A wall of text has no sections to lift. How to check: Skim the page: are there clear section headings?
  3. At least one heading is a question. Buyers ask AI questions. A heading shaped like the question makes the matching answer easy to find. How to check: Look for headings that start with how, what, why, which, can or should, or end with "?".
  4. A direct answer sits right under the H1. A short, self-contained opening paragraph is the passage most likely to be quoted. Aim for roughly 40 to 320 characters. How to check: Read only the first paragraph. Does it answer the page's question on its own?
  5. Key facts are in lists or tables. Structured blocks are extracted more reliably than the same facts buried in prose. We pass four or more list items, or any table. How to check: Are comparisons and steps laid out as lists or a table?
  6. The page is in a format answer engines cite. Guides, comparisons, reviews and “best X for Y” pages are the formats AI answers lean on most. How to check: Does the title or H1 frame the page as a guide, comparison, review or how-to?

Is the page current?

  1. The date is machine-readable. Engines favour content they can date, especially anything with prices or product names. How to check: Search the source for article:modified_time or a <time datetime> element.
  2. A date is visible on the page. It is a trust signal for readers and a freshness signal for engines. How to check: Can a visitor see a published or updated date?
  3. The page was updated in the last 18 months. Old pages get discounted, and old prices get repeated back to buyers. How to check: Check the date. If it is older than 18 months, review and update the page.

Is the page clearly yours?

  1. Your brand is named in the body copy. Engines cite names they can read, and “we” is not a name. How to check: Search the page text for your brand name.
  2. Your brand is in the title tag. It is the strongest single link between the page and your company. How to check: Look at the browser tab or the <title> in the source.
  3. There is a meta description. It states what the page answers. Aim for 50 to 170 characters. How to check: Search the source for name="description".
  4. The title is 15 to 65 characters. Long enough to be specific, short enough not to be cut off. How to check: Count the characters in the <title>.

Is the page fast?

  1. The server responds within 1.5 seconds. Slow pages get crawled less often. How to check: Run the page through PageSpeed Insights and check server response time.
  2. Largest Contentful Paint is under 2.5 seconds. It measures how fast the main content appears. Usually the hero image or a web font is the problem. How to check: PageSpeed Insights, field data section.
  3. Cumulative Layout Shift is under 0.1. Content that jumps around while loading is a poor experience. Reserve space for images and embeds. How to check: PageSpeed Insights, field data section.
  4. Interaction to Next Paint is under 200ms. Slow responses to clicks are usually long JavaScript tasks. How to check: PageSpeed Insights, field data section.

Other content signals

  1. The site has an llms.txt file. It gives AI systems a short map of your site. It costs nothing, but no major AI company has committed to reading it, so do not expect much from it alone. How to check: Open yoursite.com/llms.txt.
  2. The page links to at least three other pages on your site. Internal links are how crawlers find the rest of your content and learn which pages belong together. How to check: Count the links in the body copy that point to your own site.
  3. Images have alt text. Alt text is the only part of an image a text-based crawler can read. How to check: Search the source for <img tags without an alt attribute.

One check we score strictly

The first check fails if robots.txt blocks any AI crawler, including training-only ones like GPTBot. Blocking a training crawler is a legitimate licensing choice and does not remove you from AI answers; blocking a search crawler does. If you blocked GPTBot on purpose, that fail is expected. We explain the difference in should you block GPTBot?

What the checklist cannot cover

Every item above is about your own page. But answer engines build a lot of their answers from other people’s pages: comparison articles, review sites, forums and directories. A perfect page can still be left out if the sources an engine trusts never mention you. Getting listed and discussed in those places is the other half of the work.

Tools that do the checking for you

Questions people ask

Is there anything special I must do for Google's AI Overviews?

Google says no: there are no additional requirements or special markup for AI Overviews or AI Mode. A page must be indexed and eligible to appear in Search with a snippet. The checklist above is about making pages easy to find, read and quote, which helps everywhere.

Which items matter most?

Start with access: robots.txt, indexability and server-rendered content. If an engine cannot read the page, nothing else on the list matters. After that, a direct answer at the top and clear headings usually make the biggest difference to whether a passage can be quoted.

Will passing all 31 checks get me cited?

No checklist can promise that. These checks remove the reasons a page might be skipped. Whether you are cited also depends on what other sources say about you, which is why off-site mentions matter as much as on-page work.

How often should I run through this?

Once for every important page, then again after any redesign, CMS change or plugin update. Those are the moments a noindex tag or a robots.txt block slips in without anyone noticing.

More from the blog
Should you block GPTBot? What happens either way
A real trade with a real cost on both sides, and the robots.txt for each.
E-E-A-T when the reader is a machine
The signals a model can parse, against the ones that only move human raters.
Will AI replace SEO?
No — but the job changes, and the specific change is worth naming.

Find out where you actually stand

One real question, real AI engines, and the answer they gave — including who was named in it. No account, no card.

Check my AI visibilityRead the guides