A correction: the “22-point gap” was not a finding.

This article previously reported that the average gap between “paid-tier” and “free-tier” AI visibility was 22 points across our first 200 audits, together with a distribution, a largest-observed spread of 56 points, and a worked example. None of that was true. We had not run 200 audits, we had no such distribution, and the figures were written as illustration and then published as findings. We are replacing the article rather than deleting it, because a quiet deletion is not a correction.

What was wrong

Three separate problems, and they compound:

The sample did not exist. The numbers were presented as measured results from a body of audits we had not completed. Any reader checking our claims against our actual delivery history would have found the discrepancy immediately.

The framing was unsupportable. We described the comparison as paid subscribers versus free users. We query models through their APIs; we cannot observe what any subscription tier is served inside the ChatGPT or Gemini apps, and we never could. That claim has now been removed from the product as well as from this article.

The mechanism was over-stated. Even the parts that were directionally reasonable — that a stronger model and a faster model describe a business differently — were presented with a precision we had no basis for.

What we actually know

We know that the models inside a single assistant do not answer identically, because we query them separately and can see the difference per business. We know the direction of the effect is consistent with how these models are built: one with a smaller inference budget leans on popularity and aggregator density, one with more room engages with specific positioning.

We do not know the average size of that gap across the market. Publishing a market-wide average requires a sample we do not yet have, and when we do have one we will publish the sample size and the method alongside the number.

What your own audit reports is your own data: the citation rate for each named model, per intent, with the scan date and the models queried stated on the page. That has always been real measurement. The error was in the marketing around it, not in the audits themselves.

What we changed

The paid-versus-free framing is gone from the entire product — the site, the report, and the scoring prose. Platforms are now labelled by model name, because that is what we query. We also removed a panel on the results page that displayed an “inferred” paid-tier figure derived by adding a fixed number to a value estimated from review counts; every visitor saw the same fabricated gap presented as an insight about their business.

More broadly, we now hold a simple rule: a claim ships only if it is implemented in code and checkable against stored data. Our methodology page states what we measure and what we do not, and every audit retains its raw AI responses so you can ask us to show our work.

If you bought an audit while this article was live and want to talk about it, write to support@scoremybusiness.ai.