Quick answer: A perfect AEO audit score measures whether an engine can read you. It does not make an engine recommend you, and an unreadable site takes you out of the running before the race starts. In August 2026 my own site scored 100 out of 100 on Framer's AEO audit while my own AI Visibility Score was 18 out of 100. Both numbers are right. They measure two different gates, and the gate that decides recommendation sits on surfaces you do not own.
I run LeadJens as its own first client, Client #0, and I publish my score every month including the months it embarrasses me. That is the only reason I can write this post honestly, so here are the receipts before the argument.
What do my own numbers actually say?
My composite AI Visibility Score has moved 5, then 13, then 18 across three months, while my on-page readability sits at 100. Same business, same website, same three months.
| Month, 2026 | Composite AI Visibility Score | What moved |
|---|---|---|
| June | 5 / 100 | Baseline. Engines resolved my brand name to a different company entirely. |
| July | 13 / 100 | The site shipped. On-page and entity signals started registering. |
| August | 18 / 100 | A conversational engine finally described my business accurately, by name. |
Two honest qualifiers on that table. The Visibility Score is my own rubric, adapted from the one I run on client businesses, so treat the number as directional. The 100/100 is Framer's AEO audit, run on this site in August 2026, and it grades the page, and only the page.
Asked cold for the best service in my category, with my name never mentioned, the engines returned a competitor. That is a different question, answered from different evidence.
What does an AI readability score actually measure?
It measures retrievability: whether a machine can fetch the page, parse the structure, validate the schema, and extract a clean answer. Framer's AEO audit checks direct answers at the top of every page, structured data, question-shaped page titles, and fresh publish dates. Every one of those is a genuine prerequisite, and most agent websites fail several of them. This work matters and it has to happen.
But look at the shape of the test. It examines a page I control, against criteria I can satisfy by editing that page. Nobody else has to vouch for me anywhere in it. A tool that reads only my own claims can only grade how clearly I make them.
So why does the engine recommend someone else?
Because by the time your page is read, the shortlist already exists. None of the engines publish their retrieval architecture, so I will describe only what is observable: ask for a category recommendation and you get a handful of names, and those names correlate with independent surfaces more closely than with on-page quality. Google Business Profiles, review platforms, directories, industry coverage, community threads. The candidate's own site appears to get read after it makes the list, not in order to make it.
If you never make the list, the page never gets its turn.
The buyer behavior underneath this is measured, even where the mechanics are not. When a Google AI summary appears, people click through to a website only 8% of the time, against 15% when there is no summary (Pew Research Center, July 2025). The answer is becoming the destination, and being inside it is the whole game.
What moves you from known to recommended?
Evidence an engine can check without taking your word for it. Reviews on platforms it already reads. A complete, claimed Google Business Profile. One identical name, title, and market everywhere you appear. Presence wherever engines draw from when they assemble a shortlist in your category.
83% of consumers use Google to read reviews of local businesses, and 74% check two or more review sites before choosing (BrightLocal, 2025). The engines are reading the same layer your buyers already were, which is also the mechanism behind how agents get recommended by ChatGPT since the Yelp licensing deal. That layer lives outside your website and cannot be edited into existence.
Why is this uncomfortable to sell?
Because the half of the work I can schedule is the half that matters less. I can build the site, validate the schema, fix the entity, and make the reviews machine-readable inside a few weeks, and I can tell you the date it will be done.
The recommendation layer accumulates on its own clock. Engines re-crawl when they re-crawl, independent surfaces fill in over months, and no lever makes a third party vouch for you faster. Anyone quoting you a date for "recommended by ChatGPT" is quoting a date they invented.
I shipped my own off-site evidence layer in August. My honest expectation is that shipping it does not equal being recommended, and I will not know whether it moved until the next monthly measurement.
What should you do about it this week?
Run the two-query test on yourself before you spend anything. Open a private browser window and ask an engine two questions in this order. First, "who is [your name], real estate agent in [your market]?" Second, in a fresh session, "who is the best real estate agent in [your market]?" with your name nowhere in it.
If the first answer is accurate and the second answer does not contain you, you have the exact gap this post is about, and more website will not close it. Start with the two free fixes on the surfaces you do not own: claim and complete your Google Business Profile, and make your name, title, and market identical everywhere they appear.
Common questions
Why doesn't ChatGPT recommend me even though my website is optimized?
Because your website answers a different question than the one the engine is asking. On-page work decides whether an engine can read you once it arrives. Which businesses it arrives at is decided earlier, from third-party sources. A perfect page cannot get you onto a shortlist it was never consulted for.
What is the difference between being known by AI and being recommended by AI?
Known means the engine describes you accurately when someone types your name. Recommended means it says your name when nobody typed it. Known is mostly won on surfaces you control. Recommended turns on surfaces you do not.
Is a 100/100 AI readability score worth anything?
Yes, as a floor. It confirms your page is parseable, your schema validates, and your content is extractable. Those are real prerequisites and most agent sites fail them. It still only measures your page, which is why it can sit at 100 while an engine recommends somebody else.
How long does it take to go from known to recommended by AI?
Longer than the build, and nobody can honestly give you a date. My own composite went from 5 out of 100 in June 2026 to 13 in July to 18 in August on LeadJens's own rubric, and at the end of that I was still not recommended unprompted. That is the real shape of the curve.
What actually moves you from known to recommended?
Evidence an engine can check independently: reviews on platforms it already reads, a complete and claimed Google Business Profile, one identical name and title and market everywhere you appear, and presence in the places engines draw from when they assemble a shortlist in your category. Your website supports all of that and stands in for none of it.