The AI SEO checklist: 21 checks for ranking in Google and getting cited by LLMs
This is the checklist we run on our own properties. It covers both halves of the modern problem: ranking in Google, and being the company ChatGPT names when your buyer asks it who to hire. Work through it top to bottom. Most sites fail more than half of these.
Part 1: Entity clarity (5 checks)
Search engines and LLMs both have to answer one question before they can recommend you: what is this entity? Most B2B sites never say it plainly.
1. The first sentence of your homepage says what the company is in plain words. Not a slogan. "X is a [category] for [buyer]." If a stranger can't repeat your category after one line, neither can a language model.
2. Every major page opens by answering "what is this page about" in its first two sentences. LLMs retrieve passages, not pages. The passage has to stand alone.
3. Organization schema is on the homepage: name, URL, description, founder. This is the machine-readable version of check 1.
4. Your company name, description, and category are identical everywhere they appear: site, LinkedIn, directories, podcast bios. Inconsistent entities get low-confidence treatment, and low confidence means you don't get named.
5. An llms.txt file exists at your root, with a one-paragraph entity description and a curated list of your pages with one-line summaries. An hour of work, and it forces the discipline of checks 1 through 4.
Part 2: AI crawlability (4 checks)
6. Your content is in the HTML. If the page needs JavaScript to show its main copy, most AI crawlers see an empty div. Server-rendered or static beats client-rendered for anything you want cited.
7. robots.txt reflects a deliberate decision about AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended). Blocking them by default is a choice some sites inherit from a plugin without ever making it. If you want citations, they need access.
8. The sitemap is current and every page you care about is in it, with real lastmod dates.
9. One H1 per page, descriptive title under 60 characters, meta description that answers the query. Boring, still load-bearing.
Part 3: Content that gets cited (5 checks)
10. Each page targets one question or one keyword cluster, not five. Pages that try to answer everything get retrieved for nothing.
11. Answers come first, explanation second. Put the direct answer in the first two sentences under each H2, then elaborate. LLMs lift the tight passage; readers skim the same way.
12. Real specifics: numbers, dates, named tools, actual prices where you can, first-hand observations. Generic content is statistically invisible because a model has already seen ten thousand versions of it.
13. Topical clusters, interlinked. Three to five posts around each theme, each linking to the others and to the money page. Depth on a narrow topic beats breadth every time; it is the entire method in our AI SEO playbook.
14. Content carries a visible date and gets updated. Stale content decays in both channels.
Part 4: Schema markup (4 checks)
15. Article schema on posts, with author and publisher.
16. FAQPage schema wherever you genuinely answer questions. FAQ blocks are disproportionately retrieved by LLMs because they are pre-formatted as question and answer.
17. Service schema on service pages: what it is, who provides it, where.
18. Validate it. One malformed JSON-LD block silently disables the rest. Google's Rich Results Test takes two minutes.
Part 5: Measurement (3 checks)
19. Search Console is connected and someone reads the query report, because impressions on new queries are the earliest signal that targeting is working, months before clicks arrive.
20. AI citations are tracked: run your buyers' actual questions against ChatGPT, Perplexity, and Claude on a schedule and log whether you get named. This is the least crowded scoreboard in marketing right now.
21. You measure in business terms somewhere: queries with buying intent, form fills, booked conversations. Traffic that never becomes pipeline is a hobby.
The 10-step AI search optimization checklist for a single page
The 21 checks above are a site audit. This is the per-page version: run it on any article or service page before it ships, in order.
1. One target question or keyword cluster, stated in the title and H1 in plain words.
2. The first two sentences answer the question directly, no wind-up.
3. Every H2 is a question a buyer would actually type, or a direct answer to one.
4. At least three real specifics per section: a number, a named tool, a date, a first-hand observation.
5. Entity line present: the page says what your company is, in one sentence, somewhere above the fold.
6. Internal links to the cluster's other pages and to the money page, with descriptive anchor text.
7. Article or Service schema, and FAQPage schema if the page answers questions (it should).
8. Content is in the HTML, not injected by JavaScript, and the page is in the sitemap.
9. Title under 60 characters, meta description that answers the query, one H1.
10. A visible date, and a calendar reminder to update the page in 90 days.
Ten boxes, ten minutes. A page that passes all ten is retrievable by LLMs and rankable by Google; a page that passes five is a draft.
Where sites fail this checklist
In the audits we have run, the same three failures repeat. The entity is never stated plainly (checks 1 and 4), the content answers nothing specific (checks 10 and 11), and nobody measures the AI channel at all (check 20). Fixing just those five checks puts a site ahead of most of its market, because most of its market has not noticed the second search channel exists yet.
If you would rather have operators run the whole list, that is literally the job: AI SEO is the core of the Engine, our $1,500-a-month flat product for the MSP ecosystem.