In short: Four mistakes arrive as products. A single visibility score, a file sold as a ranking factor, a benchmark result quoted as a promise, and a tool's number treated as the target.
Each is attractive for the same reason: it converts an uncertain, slow, per-platform problem into one number that fits in a dashboard. That conversion is the error, not a simplification of it.
None of this means do not buy tools. It means know which part of what you bought is a measurement and which part is a presentation.
One Score Across Every Platform
The most widely sold and the most damaging, because it looks exactly like the metric a marketer already knows.
Assistants do not agree about which sources exist. In the 2026 citation synthesis, only about 11% of the domains cited by ChatGPT were also cited by Perplexity. Averaging across systems that disagree that much produces a number describing no platform anybody uses, and the averaging destroys the one thing you needed, which is knowing where you are weak.
The honest version is the same data unblended: appearances out of runs, per platform, per question set. It is less tidy and it tells you what to do next.
llms.txt As A Ranking Factor
A proposed convention for offering a machine-readable summary of your site. It has genuine interest behind it and no confirmed support from the major platforms.
Adding one costs an hour and harms nothing. Buying it as a ranking factor, or letting it displace the work that demonstrably matters, is ahead of the evidence. If a vendor presents it as a decided thing, that tells you how they treat evidence generally, which is more useful than the claim itself.
The general test: ask what would have to be true for the claim to be false, and how they would know. A vendor who can answer that is worth listening to on everything else. One who treats the question as hostile has told you what you needed.
The 40% Figure, Quoted As A Promise
The original GEO research reported that its methods raised visibility in generated answers by up to 40%.
That is a real published result, measured on GEO-bench, the authors' own benchmark, under their own conditions. It is a reasonable argument for taking the field seriously. It is not a forecast for your website, and it is quoted as one constantly.
The same applies to every analyst forecast about search volume: context, not evidence about you. What is GEO? covers where the number came from and what it does and does not support.
Optimising The Tool's Number
The subtlest of the four, because it happens after you have done everything else right.
A tool computes a score from a method it does not publish and may change without telling you. Once that score is the target, two things follow: a trend line that crosses a methodology change is not a trend, and work gets chosen for its effect on the score rather than on the answers.
Use a tool for what it is good at, which is running many prompts quickly. Then record the answers yourself. The thing worth owning is the record, not the automation, and a record you can hand to somebody without your login is worth more than one you cannot.
What To Actually Ask A Vendor
Four questions, and the answers are usually more informative than the demo.
Does it record the answers, or only its score for them? Can I export the raw responses? Is the result per platform or blended, and if blended, can I see it unblended? And what changed in your method in the last year, and were customers told?
A tool that stores the answers is a tool that leaves you with an asset when you stop paying for it. One that stores only its own score leaves you with nothing, which is the actual cost of the subscription.
Key Takeaways
- A single blended visibility score averages across platforms that share roughly a tenth of their cited domains, which destroys the information you needed.
- llms.txt is a proposal with no confirmed platform support. Adding one is cheap; buying it as a ranking factor is ahead of the evidence.
- The 40% figure was measured on the authors' own benchmark. It is a reason to take the field seriously, not a forecast for your site.
- Once a tool's score is the target, the method behind it can change underneath your trend line without anybody being told.
- Ask whether a tool records the answers or only its score. The record is the asset; the automation is not.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
Why is a single blended AI visibility score worse than imprecise?
- 02
What is the most useful question to ask a vendor selling an AI visibility tool?