Most articles titled “checklist” turn out to be essays with no checklist in them. This one is a list you can work through, with a way to verify each item rather than a vague instruction to improve things.
It is written for the person who owns a page and wants it cited by AI search. The background on why any of it works is in our guide to appearing in AI search results.
Work through it on one page at a time. Ten pages done properly beats a hundred done partially.
Technical checks
If retrieval systems cannot fetch and read the page, everything else on this list is decorative. Start here every time.
| # | Check | How to verify it |
|---|---|---|
| 1 | AI crawlers are not blocked in robots.txt | Open yoursite.com/robots.txt and search for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot. Confirm each block is deliberate |
| 2 | The page is indexable | Check for a noindex meta tag in the source, and confirm the URL is indexed in Search Console’s URL Inspection tool |
| 3 | Main content is in the raw HTML | View source, not the inspector, and search for a sentence from the middle of your article. If it is missing, it is rendered client side and some systems will not see it |
| 4 | The canonical URL is correct and self-referencing | Search the source for rel=“canonical” and confirm it points at the URL you want cited, without stray parameters |
| 5 | The page loads quickly on a mobile connection | Run PageSpeed Insights and check the field data. Slow pages get crawled less often, which delays everything you change |
Content structure checks
This is where most of the available gain sits, and it is all writing.
| # | Check | How to verify it |
|---|---|---|
| 6 | The question is answered in the first 60 words | Read the opening aloud. If it does not contain a specific, checkable answer, rewrite it |
| 7 | H2s are phrased as questions people actually ask | Compare your headings against Search Console query data and the People Also Ask box for your topic |
| 8 | Paragraphs are three sentences or fewer | Skim for any paragraph over four lines on desktop and split it |
| 9 | Each section stands alone | Copy one section into a blank document. If it needs earlier context to make sense, fix the opening sentence |
| 10 | Key terms are defined explicitly on first use | Search the page for each key term and confirm the first instance is followed by a definition starting with the term itself |
| 11 | Comparisons are in a table, not in prose | Look for any passage comparing three or more things across two or more dimensions and convert it |
Credibility checks
Anonymous, undated, unsourced content is the easiest thing for a model to skip.
| # | Check | How to verify it |
|---|---|---|
| 12 | A named author with relevant credentials is visible | Look at the rendered page, not the markup. A byline that only exists in JSON-LD is not a byline |
| 13 | Both a publish date and a last-updated date are shown | Confirm both appear in the visible text and match the dates in your structured data |
| 14 | Every statistic names its source and links to it | Search the page for every number and confirm each one is attributable. Delete any you cannot source |
| 15 | The page contains something first-hand | Ask what is on this page that could only have been written by someone who does this work. If the answer is nothing, that is the gap |
Structured data checks
Cheap to add, still missing from most competing pages.
| # | Check | How to verify it |
|---|---|---|
| 16 | Organization markup exists site-wide | Run the homepage through the Rich Results Test and confirm name, logo, URL and contact details |
| 17 | Article or BlogPosting markup is on content pages | Confirm headline, author, datePublished and dateModified are present and accurate |
| 18 | FAQPage markup is on any page with real questions | Validate it, and confirm every marked-up question and answer also appears visibly on the page |
| 19 | LocalBusiness or Product markup where relevant | Add for physical service areas or priced products, then validate. Skip if neither applies |
Validate everything with Google’s Rich Results Test and check the type reference at schema.org. Google’s structured data documentation explains which properties are required rather than optional.
Off-site checks
The part you cannot do on your own website, and the part that usually decides the outcome.
| # | Check | How to verify it |
|---|---|---|
| 20 | Your business details are identical everywhere | Search your business name and compare name, address, phone and description across the first two pages of results. Fix any that disagree |
| 21 | You appear in the roundups models cite | Ask an AI system your top three buying questions, note which third-party pages it cites, and check whether you are listed on them accurately |
| 22 | Independent sources mention your brand | Search your brand name excluding your own domain. If nothing credible appears, this is your highest-leverage backlog item |
Measurement checks
| # | Check | How to verify it |
|---|---|---|
| 23 | A fixed prompt set exists and is run on a schedule | Confirm you have 15 to 20 written prompts and at least two months of results, with each prompt run several times per round |
| 24 | AI referral traffic is separated in analytics | Create a segment or filter for referrals from AI platforms and confirm it is collecting data |
The Search Console performance report is your other input here. Our full method is in how to measure AI search visibility.
A worked example
We ran this list against a services page on a client site. It ranked on page one and was cited by nothing.
Six items failed. The opening 130 words were company background rather than an answer (item 6). Headings were nouns rather than questions (item 7). Three statistics had no source (item 14). There was no author or date anywhere on the page (items 12 and 13). And the pricing comparison was written as four paragraphs of prose (item 11).
None of those are technical problems. All six were fixed in an afternoon by rewriting the opening, reheading the sections, converting the pricing prose into a table, adding a byline with both dates, and either sourcing or deleting the statistics.
The pattern repeats. In most audits we run, the technical section passes and the content structure section fails. Businesses expect the problem to be a crawler setting. It is almost always the first paragraph.
How to run this on an existing page
Do it in the order the list is written, because the checks get more expensive as you go down.
- Run the technical five first. If anything fails there, fix it before touching the copy, or you will not see the effect of the copy changes.
- Rewrite the opening before anything else in the content section. It is the highest-return single edit on the page.
- Fix headings and tables in one pass. These are mechanical and quick.
- Add the credibility items. Byline, both dates, sources. Delete any statistic you cannot attribute rather than leaving it in.
- Add and validate structured data last on-page. It describes the page, so the page should be finished first.
- Record a baseline before you publish the changes, then re-run your prompt set a month later. Without the before, you cannot tell whether any of it worked.
If you would rather not do this yourself, send us your five most important pages. We will run all 24 checks and come back with the failures ranked by what will move first.