Half your pages get the change. Half don't. Eight weeks later the difference between them is your answer — not an opinion, not a score.
Half change, half don't. The gap between the two groups is the only thing that can tell you whether the change did anything.
BotPass is a GEO platform that proves which on-page changes make AI assistants cite you, by running a randomised controlled experiment on your own site. A change is applied to a randomly drawn half of your eligible pages; the other half is left untouched as a control. After eight weeks we compare how often each group is cited and report the difference with its confidence interval.
Changes that clear the bar enter a shared playbook and are applied automatically to every customer with that page type. Changes that don't are discarded — and we tell you.
They ask the AIs questions and watch what comes back. That tells you what happened. It cannot tell you what caused it, because nothing on your site changed on purpose.
With nothing to compare against, any lift can be attributed to any cause. That is not measurement. It is a story that happens to fit.
All of GEO today is correlational. We built the control group.
Five steps. Nothing enters the playbook without going through all of them.
One change, from a closed catalogue. Never a free-text idea.
Your pages are drawn at random into treated and control.
The plugin changes the treated half only. Always reversible.
We ask the AIs directly and log every crawler that returns.
The difference between groups, with its margin of error.
All three are legitimate. They answer different questions, and only one of them answers why.
| BotPass | AI visibility tool | GEO agency | |
|---|---|---|---|
| Tells you what happened | Yes | Yes | Yes |
| Tells you what caused it | Yes — control group | No | No |
| Changes your pages for you | Yes, automatically | No | Yes, by hand |
| Every change reversible and logged | Yes | n/a | Depends |
| Gets better from other clients' results | Yes — shared playbook | No | No |
| Will tell you a change did nothing | Yes, in writing | n/a | Rarely |
| Cost of entry | $39/mo | $100–500/mo | $2,000+/mo |
Competitor price ranges are typical published list prices, not quotes. Keep your visibility tool — we can read its data as a measurement source.
It's called the extraction ratio, and it can only be computed from your own server logs. Here is the industry picture. Yours will be different, and finding out costs nothing.
Pages read per referred visit. Third-party published figures, July 2026. Bars use a logarithmic scale — the real spread is far wider than it looks.
Install, paste the key, done. Any theme. Sits alongside Yoast or RankMath. Works with WooCommerce, WPML and Polylang.
REST, no plugin. Shopify, Magento, PrestaShop, headless or custom. Your team connects it once.
One click undoes any change, with a record of what was applied, where and when. Nobody from our team gets an account on your site.
We do not cloak. The parity report documents it URL by URL, dated, for your SEO lead and your legal team.
Invisible changes apply themselves. Anything that alters visible text waits for your sign-off.
Never your whole site at once. That is not caution — it is what makes the result mean anything.
Nobody controls what a model answers. Anyone who guarantees it is lying, and you'll find out in six months.
An AI takes weeks to come back and re-read your site. Before that there is nothing to measure.
When an experiment comes back flat we say so, in the same detail. You just saved the budget you were about to spend there.
We sell certainty, not promises.
Per domain. The diagnostic is free and needs no card.
The three-month minimum is not a lock-in clause: one experiment does not fit into less time. Full terms in our Terms of Service.
Those tools measure from the outside: they query the models and report what came back. They tell you what happened and they do it well. They cannot tell you why, because nothing on your site changed on purpose. We change a randomly drawn half of your pages and compare against the untouched half — that comparison is what licenses a causal claim. Keep your tool; we can read its data as a measurement source.
Six to eight weeks from the moment the experiment starts. That is how long an AI takes to come back, re-read your pages and reflect them in answers. Anyone promising results sooner is not measuring anything.
No. We serve exactly the same content to people and to machines, and we issue a parity report, by URL and date, that proves it to your SEO team and your legal department.
We tell you in the same detail as if it had worked, and we pick the next hypothesis. Knowing something does not work saves you the budget you were about to spend on it. That is part of the product, not a failure of it.
Nobody from our team. A plugin you install, on your own infrastructure with credentials you control, pulls a task list and applies changes only to the pages in the experiment. Every change comes from a closed catalogue, and every one is reversible.
Fewer than there will be a year from now, and we will tell you the exact number across the network before you sign anything. The catalogue starts deliberately narrow: we would rather apply five measured things than thirty that sound reasonable.
It connects over the REST API and needs no plugin. Shopify, Magento, PrestaShop, headless or custom builds all work the same way; the only difference is that your technical team does the integration once instead of installing from a dashboard.
No. We do not modify your template or your styles, and we coexist with Yoast, RankMath and the rest. If something does not convince you, any change reverts with one click.
What is shared is the aggregate measured effect of each change — the intervention, the page type, the effect and its confidence interval. Never your content, your traffic, your keywords, your domain or your company name. It is written into the contract from day one; see the Data Processing Agreement.
The diagnostic is free, measures over 28 days and commits you to nothing. Then you decide whether you want to know what actually works on your site.