Measured on ChatGPT · Perplexity · Gemini · Google AI Overviews
Your buyers ask ChatGPT what to buy. In 48 hours, 150 real answers show how often it names you. Then we publish the pages that move that number.
What we believe
THE OLD GAME
Position
Being one of Google's ten blue links.
THE NEW GAME
Frequency
Being inside the answer. Half of B2B software buyers now start their research in an AI chatbot, not in Google (G2, Answer Economy report, Mar 2026).
A chatbot has no fixed list. Every answer is a fresh draw. If you still track position, you're measuring a game your buyer already left.
What's at stake
If you're not, you never find out you lost. In your analytics it shows up as "direct traffic", when it shows up at all.
And the answer doesn't come only from your site. It comes from third-party listicles, review sites and communities.
Source that weighs most on a B2B buyer's shortlist
Share of B2B buyers. G2, Buyer Behavior Report, Jun 2026
Why what exists today falls short
01
Position in AI doesn’t hold still. Ask the same question twice, hours apart on the same day, and the brand moves 1.7 positions on average (Windsor.ai, Sep 2026, vendor’s own test of 30 prompts). A dashboard that says “you’re #3” is a photo of a point that already moved.
02
The API isn’t your buyer’s screen. Answers pulled through the API and answers shown in the chat window differ in the brands they name and the sources they cite. Tune for the API and you tune for an answer nobody reads.
03
Knowing you’re missing doesn’t put you in. The bottleneck was never the chart. It’s the page nobody published.
04
Its visibility score has no published formula. Its case studies show percent lifts with no baseline. The price comes after a sales demo. You’re asked to trust the number before you can check it.
05
It chases search volume. But for each question, AI runs 5 to 11 sub-searches, and over 95% of them show no volume in any keyword tool (Ahrefs, Mar 2026). Many also sell llms.txt and schema as GEO; Ahrefs’ 2026 studies found no effect of either on citations (May and Jun 2026).
06
It costs nothing this month. The cost is the demo that never got booked, and that never shows up in a report.
Check us with your own AI
You already work with Claude or ChatGPT. Use it to test the problem before you talk to us, and to audit our numbers after.
BEFORE · 2 MINUTES
Open 10 new chats in ChatGPT and paste this in each one. Count how many shortlists name you.
What are the best [your category] tools for [your buyer]? Give me a shortlist of 5 with one line on each.
That count out of 10 is your inclusion rate on one question, one engine. Run it again tomorrow and watch it move. That swing is why we use 10 runs per question and 1,200 answers.
AFTER · EVERY MEASUREMENT
Every measurement ships as files, not a dashboard login: every answer with its engine, date and login state, plus the method. Drop them into your agent and ask:
Recompute the inclusion rate per engine from these answers. Give n and a 95% interval for each. Flag any number in the report that doesn't match.
If your agent finds a number we can't back, we fix the report.
How to vet any GEO vendor, us included
| QUESTION | OUR ANSWER |
|---|---|
| 1. Do you measure position or frequency? If position, stop here. | Frequency |
| 2. Do you collect through the API or through the interface buyers use? | The interface |
| 3. How many times do you repeat each question? What's the margin of error? | 10 times, with a margin on every number |
| 4. Can I see the question list before I sign? | Yes. You approve all 30 |
| 5. What do you publish, and in how many days? | 4 new pages and 3 revised, in 45 days |
| 6. Can I see real answers before a sales call? | Yes. 150 in 48 hours, no call |
How we measure
| WHAT | WHY | WHAT CHANGES FOR YOU |
|---|---|---|
| 30 questions your buyer asks before buying: alternatives, comparisons, vertical, integrations, pain | That's where the decision happens | You measure what drives demos, not visits |
| 10 runs per question | One answer is a coin toss. Ten become a frequency | The number stops swinging with luck |
| 4 engines (ChatGPT, Perplexity, Gemini, Google AI Overviews), captured in the interface | API and interface diverge | What you see is what your buyer sees |
| A margin of error on every number | AI varies on its own | If a change fits inside the margin, we tell you it didn't change |
The limit we declare: AI personalizes answers for signed-in users. No outside measurement sees that fully, ours included. We measure signed in and signed out, and report the two separately.
How we work
Week 1
Question map: category, vertical, competitor and switching reason, pain, integrations. For each one, the sub-searches AI runs.
You get: the 30 questions, approved by you.
Weeks 1–2
Baseline: 1,200 answers.
You get: your inclusion rate by engine, and who shows up in your place.
Week 2
Source diagnosis: what AI cites when it doesn’t cite you.
You get: for every question you lose, the source replacing you: a page you’re missing, a third-party listicle, a review site.
Weeks 3–5
Execution: every page starts as a table row and passes automated checks before going live.
You get: 4 new pages, 3 revised, 3 placement outreaches made.
Weeks 6–7
Retest: same questions, same engines, same time of day.
You get: before and after, with the margin. What moved, what didn’t, the next step.
The scoreboard
| Category | Questions | Answers | Inclusion before | Inclusion after | 95% CI | Days |
|---|---|---|---|---|---|---|
| No sprint measured yet. The first row goes in when the first sprint closes. | ||||||
A lift with no baseline, no sample size and no date isn't a result. Every row here will have all three. Until the first one lands, the most honest number we can show you is yours.
Measure my inclusion rate →A position
In AI answers, it doesn't hold.
A 0–100 score
Nobody can audit it.
llms.txt and schema as GEO
The 2026 studies found no effect.
A demo before the data
You see real answers first. Talking to us is optional.
A promise without proof
Nothing the margin of error can't back.
The offer
On day 45 you open a report with the same 30 questions and the same 1,200 answers, and see how many now name you. And why, source by source.
The method is on this page. The scoreboard is public. The questions are yours.
Fixed price, in USD.
You get the number after your free measurement, along with what the sprint would target in your category.
Includes: the 30-question map, 1,200 baseline answers, source-by-source diagnosis, 4 new pages, 3 revised pages, 3 placement outreaches, and the day-45 retest.
Start with the free measurement →It's for you if
It's not for you if
The interface. It’s what your buyer sees, and the two diverge too much for one to stand in for the other.
Because the same question almost never returns the same list. With one run, you’re measuring luck.
No. Nobody controls the AI’s answer; be wary of anyone who says they do. What we commit to is the protocol, the published pages and the measured number.
The report shows why, source by source, and what we’d change next.
It’s a fixed price in USD. We quote it after your free measurement, once we know how many questions you’re losing and to whom.
Yes. Every measurement ships as files: each answer with its engine, date and login state, plus the method. Your agent can recompute every number in the report.
No. We need publishing access, or someone who approves drafts once a week.
We measure on day 45. Some pages get cited earlier; others wait on indexing. The report keeps the two apart.
150 real answers about your category, in 48 hours. Free.