The AI SEO Foundation Checklist

This is the 42-point list I run on every site before I write a single line of new content. It is the same sequence I walk through in the video. Work top to bottom. If AI crawlers cannot reach you, nothing further down the list matters.

42 checks 7 stages 1,000+ sites this has been run on Last updated 14 July 2026

0 of 42 done

Let the AI crawlers in

You cannot be cited by a model that never fetched your page. Most sites I audit are quietly blocking at least one major AI user agent without knowing it.

0 / 6

Allow the AI user agents you want citations fromDo firstOpen robots.txt and check each one deliberately. Blocking a training crawler and blocking a live retrieval crawler are two different business decisions. Bot names change often, so verify the current user agent strings before you edit.
Confirm your CDN or WAF is not blocking them anywayCloudflare, Sucuri and most managed hosts ship bot-fighting rules that are on by default. A permissive robots.txt means nothing if the edge returns a 403 before the request ever reaches your server.
Serve your primary content in the initial HTMLCurl your own page and read what comes back. If the answer, the pricing or the product copy only appears after JavaScript runs, assume the model never saw it.
Publish an llms.txt at your rootNo assistant guarantees it reads this today. It costs an hour, forces you to name your most important pages, and is trivial to remove if the standard dies.
Find the AI crawlers in your raw access logsThis is the only ground truth. Grep the log for the AI user agents and count the hits and the status codes. If you see zeros or 403s, go back to the first two checks.
Keep the XML sitemap honestEvery live URL in, every dead and redirected URL out, and lastmod dates that reflect real edits. A stale sitemap teaches crawlers to distrust your freshness signals.

Define the entity

A model has to be sure who you are before it will put your name in an answer. Ambiguity is the single most common reason a brand gets left out.

0 / 6

Ship Organization or Person schema with a stable @idDo firstOne canonical node, one permanent @id, referenced from every page on the site. This is the anchor everything else hangs off.
Build a sameAs array of profiles you actually controlLinkedIn, the platforms you sell on, your socials, your company registration. Every URL must resolve and must genuinely be you. A fabricated Wikidata ID in a sameAs array is worse than no sameAs at all.
Make name, address and phone byte-identical everywhereSuite 4 in one place and Ste. 4 in another is enough to split you into two weakly-supported entities in a model's head. Pick one string and enforce it.
Write an About page that says plainly what you doFounded, based, serves, specialises in. Boring declarative sentences. This is the page a model reads to decide whether you belong in the shortlist for a category.
Give every author a real bio page with credentialsA byline that links nowhere carries no weight. A byline that links to a page with experience, credentials and a sameAs array carries a lot.
Disambiguate yourself from your name-twinsSearch your own brand name and see who else you collide with. If another entity dominates, your job is to attach a distinct qualifier to yourself consistently: the category, the city, the niche.

Ship structured data that is true

Schema is not a ranking cheat code. It is a machine-readable restatement of what your page already says. Inflate it and you invite a manual action.

0 / 6

Use one @graph block per page, not five loose scriptsA single graph lets nodes reference each other cleanly. Scattered blocks duplicate entities and produce contradictions parsers have to guess their way out of.
Match the type to the job the page doesArticle for posts, Product for products, FAQPage only where visible questions and answers exist, HowTo only for a real ordered process, LocalBusiness only if you have a real address.
Reference shared nodes by @id instead of retyping themYour Person or Organization node is defined once. Every other page points at it. That is what turns a pile of pages into one entity.
Only mark up ratings and reviews that genuinely existDo firstAn AggregateRating with an invented review count is a schema violation, not a growth hack. I see it constantly on competitor sites. It is a liability with a delayed fuse.
Validate every template programmatically before it shipsNot eyeballed. Parsed. One broken quote in JSON-LD silently voids the entire block, and you will not see the loss in any report.
Check the schema mirrors the visible pageIf the FAQ answers exist only in the markup and not on the page, you have built a cloaking problem, not an optimisation.

Make the content extractable

A model does not rank your page. It lifts a claim out of it. Write so that a single clean, quotable claim is easy to find.

0 / 7

Answer the question in the first 40 to 60 wordsDo firstDirectly under the heading, before any context or backstory. That opening block is the piece most likely to be quoted verbatim. Everything else is supporting evidence.
Phrase H2s the way people actually askHow much does X cost beats Pricing. The heading is the retrieval key. Match the question, not the category label.
Put comparisons in real HTML tablesAssistants lift tables almost intact when answering X vs Y queries. A comparison buried in three paragraphs of prose gets skipped for a competitor who used a table.
Define your terms in one clean sentenceTerm, is, definition. No wind-up. Definitional sentences are the highest-value extraction target on the internet right now and almost nobody writes them properly.
Attribute every statistic to a named, dated sourceAn unattributed number is a liability a model will not repeat. A cited number with a date and a link is a claim it can safely stand behind.
Add something a model cannot paraphrase from elsewhereYour own test results, your own client numbers, your own screenshots, your own opinion with your name on it. Rewritten consensus gets no citation because the consensus already exists.
One primary topic per URLA page that half-covers four subjects is the best-ranked answer to none of them. Split it.

Wire the site together

Internal links are how you tell a crawler which of your pages is the authority on a topic. Orphaned content is invisible content.

0 / 6

Give each core topic one hub pageThe definitive page you want cited. Everything else on that topic exists to support it and links up to it.
Link spokes up to the hub and sideways to siblingsA cluster with no lateral links is just a list. The sideways links are what signal the topic has depth behind it.
Kill the orphansDo firstCrawl your own site and list every page with zero internal inbound links. On most blogs this is the majority of posts. Fix it with cornerstone links from the homepage and topic hubs.
Use anchor text that describes the destinationRead more tells a crawler nothing. The anchor is a label, so label the page.
Add breadcrumbs in the markup and the interfaceBreadcrumbList schema plus a visible trail. Cheap to add, and it tells a machine exactly where the page sits in your hierarchy.
Keep money pages within three clicks of homeDepth is a proxy for importance. If your best page is five clicks deep, you have told the crawler it does not matter much.

Earn the off-site signals

Assistants overwhelmingly recommend brands that other sources already vouch for. What is said about you off your domain is doing more work than your homepage is.

0 / 6

Get into the listicles the models already quoteDo firstAsk an assistant for the best providers in your category, then look at what it cites. Those exact pages are your target list. Pitch them, contribute to them, get added to them.
Be genuinely useful where your buyers already talkReddit, Quora and niche forums are heavily retrieved. Turning up to drop links gets you removed. Answering hard questions properly, under a real profile, gets you quoted.
Claim and complete the profiles in your categoryReview sites, marketplaces, industry directories. Half-finished profiles are an entity liability. Completed ones are corroboration.
Publish something original enough to be citedA survey, a benchmark, a teardown, a dataset. Original numbers are the most reliable way to become the primary source instead of one of forty paraphrases.
Make your unlinked mentions consistentBrand mentions count even without a link, but only if the wording is stable enough to be recognised as the same entity every time.
Align third-party profile copy with your site copyIf your site says AI SEO and your marketplace profile says web design, you have given the model two conflicting stories and it will trust neither.

Measure what AI search actually sends

If you cannot see it, you cannot defend the budget. Set this up before you start the work, not after.

0 / 5

Segment AI assistant referrals in GA4Do firstBuild a channel group for referrals from the assistant domains. Out of the box they get dumped into generic referral traffic and you never see them.
Track AI crawler hits from raw server logsReferrals tell you who clicked through. Logs tell you who is reading you at all. The gap between the two numbers is your citation problem.
Run a fixed prompt set every monthTwenty real buyer questions. Same wording, same assistants, same day of the month. Log who gets named and who gets cited. That is your rank tracker now.
Watch branded search volume as a leading indicatorAI recommendations often produce a branded search rather than a click. Branded queries climbing while referrals stay flat usually means the citations are landing.
Re-run this checklist every quarterCrawler names, retrieval behaviour and schema support all move faster than your CMS does. What passed in January will not necessarily pass in June.

Questions I get asked

Before you start

Is AI SEO different from normal SEO?

The foundation overlaps heavily. Crawlability, clean structure and real authority still decide everything. What changes is the output: a model has to extract a specific claim from your page and attribute it to your brand, so answer-first writing, accurate schema and consistent entity signals matter far more than keyword density ever did.

How long does an AI SEO foundation take to build?

Crawler access and schema are usually done in a week. Entity consistency and internal linking take two to four weeks on a mid-sized site. Off-site citation signals are the slow part and typically take three to six months to move.

Do I need an llms.txt file?

No assistant guarantees it reads llms.txt today. It costs almost nothing to publish, it forces you to name your most important pages, and it is trivial to remove later if the standard dies. That risk profile makes it worth doing.

How do I know if AI search is sending me traffic?

Build a GA4 segment for referrals from the assistant domains, then cross-check it against raw server logs for AI crawler user agents. Referral data tells you who arrived. Log data tells you who is reading you in the first place.

Want me to run it for you?

Send me your URL and I will run this checklist against your site and tell you exactly where it breaks. No cost, no pitch deck.

Get a free AI SEO audit Hire me on Upwork