Home/Learn/AI SEO analyzer

Get your pages picked by AI assistants

An AI SEO analyzer checks whether a page is crawlable, specific and easy to cite. Use those checks to fix the page, then ask buyers' questions on the assistants and count the visits they send. The page score alone cannot show whether the company appears in answers or whether referred visitors become inquiries.

Updated November 2026

The analyzer returns a score out of 100 and a list of changes for the URL you entered. A file may be missing; a paragraph may need more detail. You can work through that list and still be unsure whether an assistant would recommend the company to a buyer.

An "AI rank checker" faces the same distinction. AI answers depend on the question and the assistant, so there is no single fixed rank to report. A useful review separates what can be inspected on the page from what needs to be observed in answers and visits.

An AI SEO analyzer checks three things on a page

Use the analyzer to inspect whether an assistant can fetch the page, understand its specific claims and find material it can cite. These are concrete parts of the page to examine and improve.

Crawlability concerns access. The check asks whether the relevant crawler can fetch the public page. A page meant to inform buyers needs to be available to the systems reading it. The guide to robots.txt and AI crawlers covers the crawl rules in more detail.

For specificity, read what the page says about the company: the work it does, the people it helps and the facts behind its claims. An assistant answering a particular question needs that detail. A sentence about helping businesses leaves it with little to explain.

Citation checks concern how the information is presented. Structured data, a markdown copy and an llms.txt file are among the items examined in this work. They give the reviewer concrete files and page versions to inspect alongside the public text.

Read the findings by the object they identify. A crawl-rule finding calls for a file review. A vague service description calls for writing based on the company's actual work. The analyzer's list can help locate both, but the resulting assignments involve different kinds of judgment.

The guide on getting ChatGPT to recommend your business connects these page properties with measuring qualified visits. That is the next part of the job after the page checks: examining how assistants use the information and what people do when they arrive.

A high score cannot show whether assistants name you

An assistant can read a company's website and still leave the company out of its answer. A page score does not capture that decision for the buyer question you care about.

In an August 2026 observation, an assistant found a startup's site, read its information and then omitted the company. The problem was no longer whether the page could be opened. The information had to make a persuasive case for including the company in that answer.

A useful test therefore starts with a buyer's question that does not name the product. Asking an assistant directly about a company gives it the subject of the answer. Asking for help with the buyer's problem shows whether the assistant introduces the company while considering the available options.

Run those questions across ChatGPT, Perplexity and Gemini, retaining the answers. The assistant and question both matter when interpreting what appears. A company may be relevant to one question and absent from another, so the result needs that context beside it.

Read what the answer says about the company, as well as whether it names it. The description can show which parts of the page were useful and which details were missing. Use those details to identify the next writing assignment.

When reviewing a report, locate both records. The page checks tell you which files and descriptions were inspected. The answers tell you what the assistants said to the chosen buyer questions. Each can identify work, but they describe different observations.

Count assistant visits, then see who gets in touch

Connect visits referred by assistants with the inquiries those visitors make, while keeping automated page requests separate. That shows which visits become a conversation with a potential buyer.

An agent request and a person arriving from an assistant tell you different things. The request shows that a system fetched a page. A referred visitor shows that someone followed a route to the site. Inquiry tracking adds what that person did after arriving.

Testing can complicate the count. On our own projects, an agent's browser testing once appeared as traffic. A report that included those visits could read our own checks as interest from outside the company.

Separate agent and testing activity before using visits to judge the page work. Keep the request record for understanding what agents fetch, and use the visitor record to examine people coming to the site. Each record has a useful job when its meaning stays clear.

Then connect the visitor's arrival with the action the page offers. Conversion tracking should show whether the person went on to make an inquiry. A referral count without that next action leaves the company unsure whether assistants are sending people interested in its work.

This is why the measurement extends beyond mentions. An answer can name the company while the report still has no evidence of an inquiry. Reading the answer, the referred visit and the contact separately gives the founder a clearer view of which part of the route needs attention.

A person must write the facts an analyzer flags

Making a vague page useful requires someone who knows the company's work to supply the facts behind its claims. The analyzer can point to missing detail; the writer has to establish and explain it.

A service description should say what the company does for a buyer in a recognizable situation. That requires choosing details from the actual work and connecting them to the question the page answers. Adding a generic paragraph about quality gives the reader little more to judge.

This matters for B2B trust as well as assistant answers. Our writing experience points to a loss of trust when a page reads as generic AI output. Specific facts and permitted examples help explain the work in terms a prospective buyer can assess.

Fixing the analyzer's list yourself gives you direct control over those choices. It suits a founder who knows the facts and wants to write or direct the changes. The assignment still includes returning to the answers and checking whether referred visits lead to inquiries.

A tracking tool that counts mentions supplies repeated observations over time. It can help locate questions where the company is absent or its presence changes. Someone still needs to read the answers behind those counts, write the missing detail and inspect the visitor response.

With people handling both the revision and the review, the flagged page becomes a writing assignment. They add the company's facts, return to the same buyer questions and inspect the referrals after the change. The next revision can then respond to what those checks show.

Before choosing help, ask who will write the details missing from the page. Then ask who will read the answers after publication and check whether referred visitors make inquiries. Those are separate jobs to include, even when a report puts them beside the same score.

A coding agent can run the checks on your site

On a site whose files you control, Codex or Claude can help inspect and change the files an analyzer flags. Crawl rules, structured data and text copies give the agent concrete work to perform.

If you built the website, you already have files for a coding agent to work on. It can inspect the crawl rules or a text copy, edit what needs changing and examine the new version. You still need someone to decide what the company can truthfully say in that version.

An open-source marketing system for coding agents also exists for founders who want to run this work themselves. That is an available route for someone who wants to maintain the checks with their site's code. It still requires attention to how the checks and writing behave.

In our experience, running this work through Codex or Claude requires repeated tuning. The file edit is only part of it. Someone also needs to examine the writing instructions, keep the company facts accurate and read the answers returned after changes to the site.

Separate those responsibilities when deciding how much to run yourself. A coding agent can handle a file change and help repeat an inspection. The person directing it needs to judge whether the page now says something specific and true about the company.

Maintaining this yourself means returning to it when the site changes. A founder can direct those checks through the code; people handling the writing can include them with page revisions. After either route, the question is what the assistants now answer and whether their visitors contact the company.

Example: opening a site to assistants and logging visits

As of November 2026, our own site provides text versions and question-based access alongside a record of agent requests.

Site featureWhat it makes available
Markdown copy of every public pageA text version of the page
/llms.txtAn index of the site
/ask?q=The page that best answers the submitted question
Agent-request recordingRequests with emails, tokens and keys removed

These features make the pages available in forms agents can read and show which requests they make. Visitor and inquiry tracking supplies the separate view of people arriving from assistants.

Fix the analyzer's list, then test the answers

  • Inspect the flagged files

    Run the analyzer and address its crawl-rule, structured-data, llms.txt and markdown-copy findings against the public pages they describe.

  • Replace vague descriptions

    Write the company's own facts into the flagged pages so a buyer can understand the work and whom it helps.

  • Ask unprompted buyer questions

    Test questions that do not name the company on ChatGPT, Perplexity and Gemini, then read the descriptions and recommendations returned.

  • Follow assistant referrals

    Separate agent and test traffic, then count referred visitors who make an inquiry to see whether the visits lead to contact.

Test whether assistants send you buyers

Talk to Tin about the pages your analyzer flags and the answers your buyers receive. Our growth experiments connect factual page revisions with repeated answer checks and inquiry tracking, so the conversation can address what needs work next.

Book a call