FREEFirst project with us? Your landing page design is on the house.See the offer

Chapter 6 of 6

Measuring it, and what nobody can promise

Three layers of measurement, a monthly prompt panel, honest limits and a one-page audit.

6 min readUpdated 2026-11-07
All chaptersChapter 6 of 6
  1. 01What AEO is and what changes
  2. 02Answer-first copy
  3. 03Schema per page type
  4. 04llms.txt and AI crawlers
  5. 05FAQ, comparison and glossary pages
  6. 06Measuring it, and what nobody can promise

The short answer

Measure what you control every month: access, indexing, page checks and copy coverage. Sample what you do not control: whether AI tools name you and how accurately. Count what reaches you: referrals and leads who say they found you through an AI tool. Treat all of it as a direction of travel. Nobody can promise a citation, and no single score can tell you that you will get one.

Three layers of measurement

LayerWhat to trackWhereHow often
Inputs you controlTechnical checks passed, pages with an answer block, valid schema, crawler accessAEO score checker, AI crawler checker, Rich Results TestMonthly, and after any release
Search presenceIndexed pages, impressions and clicks for your question queriesGoogle Search ConsoleMonthly
Answers you can sampleNamed or not, cited or not, accurate or notYour own prompt panel, belowMonthly
Business signalsAI referrals, form answers, sales notesAnalytics and your lead sheetMonthly

The score checker fetches one page and runs 12 technical checks, including title and description length, one h1, canonical link, JSON-LD, AI crawler rules, llms.txt and a sitemap. The score is the number passed. It is a checklist of signals, and it does not predict rankings or citations. Use it to find gaps, then stop looking at it.

What Google reports

Google says clicks from AI features count towards overall Search traffic and appear in the Performance report under the Web search type. In practice, watch your total clicks and your question-style queries together. Check Google's current docs for any newer breakdown before you build a report on this.

If a page should appear and does not, work through the basics: use URL Inspection to confirm indexing and what Googlebot sees, and check for nosnippet or noindex. If you change those, Google says recrawling can take from several days to several months.

Build a prompt panel

This is manual, and it is the most useful thing on this page. A prompt panel is a fixed list of buyer questions that you ask the same tools each month.

  1. Write 10 to 20 questions from sales calls. Mix three kinds: about the category ("how do I move from WordPress to Webflow"), about options ("what should I look for in a Webflow agency") and about you ("what does [your company] charge").
  2. Choose your tools. Pick the ones your buyers use, such as Google's AI features, ChatGPT, Claude and Perplexity.
  3. Log each answer in a sheet. Columns: date, tool, question, were you named, were you linked, which URL, were the facts right, which sources were cited instead.
  4. Act on it. A wrong fact goes to the copy or to a third-party correction. A rival's page that is often cited is worth reading to learn what it says clearly that you do not.

Be honest about the limits. Answers can change between runs and between people, so a single result proves little. Look for patterns over months, and do not celebrate or panic over one screenshot. See AI visibility and AI citation for the terms.

Count what reaches you

  • Referrals. In your analytics, filter sources for domains such as chatgpt.com, perplexity.ai, claude.ai and gemini.google.com. Expect this to undercount, because a person may read an answer and then search for your name or type your address.
  • Ask people. Add "How did you hear about us?" to your form with a free text box, and have sales note the answer on calls. If someone says "ChatGPT told me", that counts, even when analytics shows direct traffic.
  • Branded search. Watch searches for your name in Search Console. It has many causes, so use it as context and not as proof.

Our post on marketing site tracking covers the setup behind these numbers.

What nobody can promise

Be direct about this with your team, and expect the same from anyone you hire.

  • A citation. Each company decides how its product chooses sources. Google says indexing and serving are not guaranteed, even when requirements are met.
  • A rich result. Google says valid markup does not guarantee one.
  • Accuracy. A tool can get your facts wrong even on a clear page. See AI hallucination.
  • Lasting placement. Products and their sources change.
  • A benefit from llms.txt. It is a proposal, and Google says it is not needed.
  • Traffic. An answer may satisfy the reader with no click at all.

Red flags when buying this work: guaranteed citations, a secret formula, a score presented as a ranking factor, pages generated in bulk with no human check, hidden text for AI, or fake reviews. Ask instead: what will you change, how will we check it, and what will you do if the sampled answers do not move?

The one-page audit

Print this, and mark each line yes, no or not relevant. Start with your five main pages.

Reachable

  • A written policy for training, search and user crawlers exists.
  • robots.txt follows it, and the CDN or firewall agrees.
  • The main pages are indexed and allowed a snippet.
  • Important text is in the HTML source.

Readable

  • The page's first block says what it is, who it is for and the key fact.
  • The first sentence under each main heading passes the lift test.
  • Headings use the buyer's words.
  • Prices, times and limits appear as text, and you are happy to be quoted on them.

Described

  • Organization and BreadcrumbList are present and valid.
  • Each page type has the schema from the table in chapter 3, and it matches the visible page.
  • Articles name a real author, with honest dates.
  • Facts come from one source, and third-party profiles agree with it.
  • llms.txt is current, or you have decided not to have one.

Quotable

  • A visible FAQ answers real questions, including the awkward ones.
  • Comparison pages open with a fair verdict and rival facts you read at the source.
  • Glossary entries open with a definition, and exist only where you can add something.

Measured

  • A prompt panel is logged monthly, with a named owner.
  • AI referrals and "how did you hear about us" answers are reviewed monthly.
  • The audit runs again after any redesign or platform move.

What to fix first

Fix in this order, and stop when the next item is not worth your time.

  1. Anything that blocks access or indexing.
  2. Wrong facts, on your site or elsewhere.
  3. Answers on your five main pages.
  4. Schema on your templates.
  5. Missing FAQ, comparison or glossary pages your buyers ask about.
  6. llms.txt and other extras.

Then get on with running the site. If you would like us to run the audit or the fixes, talk to us or look at our SEO, AEO and GEO service.

Takeaways

What to remember from chapter 6.

  • Measure inputs, sample answers and count what reaches you.
  • Google reports AI feature clicks within the Web search type in Search Console.
  • A prompt panel shows patterns over months. One result proves little.
  • Be wary of anyone who guarantees citations.

Keep a copy

Take the whole guide
as a PDF.

It all stays free to read here. If you want a copy for your team, leave your email and the download link shows up right away.

Nothing gets emailed to you, the link appears on this page. We do not send marketing email unless you ask. Privacy policy

Your turn

Let's kook
what you krave.

Hi Kookies, I need a built on . I'd love to launch . Reach me at
or book a callSent. We reply within one working day.