Prisma shows how AI assistants find, understand and recommend your business, and is built to prove what changes their answer.
The shortlist is written before anyone visits your site. If the AI leaves you out, the buyer may never know you exist.
of Google visits include a click on a traditional result with an AI summary, versus without one (US).
Pew Research, 2025weekly active ChatGPT users.
OpenAI, Feb 2026of B2B software buyers now begin their purchasing research in an AI chatbot.
G2, Apr 2026AI-referred visits to US retail sites, May 2026 versus May 2025.
Adobe, Jun 2026of Brazilian internet users already use generative AI, about 50 million people.
Cetic.br, 2025The AI knows you exist
It links to your pages
It tells the buyer to choose
The buyer follows the advice
Illustration of the exposure ladder Prisma measures, step by step.
An AI can know your brand, cite your website and still recommend a competitor. Visibility dashboards count mentions. The sale is decided one step later.
buyer conversations advised buying the product we tracked right away. It was named in only 3.
Prisma lab, 10 Sep 2026 · four AI setups · exploratory studyproducts named for the same question in one run, then in the next. Same account, same mode, same day.
Prisma lab, 11 Sep 2026 · exploratory studyHow an AI decides who to recommend is not a total mystery. We map what the research we read and check, and the answers we capture, show.
Press, encyclopedias, forums, video, reviews and government sites. Your website is one voice among many.
of links cited by ChatGPT, Claude and Gemini come from earned media.
Muck Rack, May 2026 · 25M+ linksmonitored Brazilian brands had AI answers citing Reclame Aqui.
Promptado, Jan 2026 · vendor dataEach assistant leans on different sources. Watch the map switch engine, or pick one yourself.
of cited domains are shared by ChatGPT and Perplexity.
Profound, Jul 2025 · 100,000 promptsof citations to ChatGPT's ten most-cited sites go to Wikipedia. On Perplexity, Reddit gets 46.7%.
Profound, 2025 · 680M citationsAsk twice and the shortlist moves. Ask as someone else and it moves again. One check tells you very little.
chance of getting the same list of brands twice from ChatGPT or Google AI.
SparkToro, Jan 2026 · 2,961 runsbrands shared by answers for the same buyer profile, versus different profiles.
Profound, Sep 2026 · 71,147 answersTechnical tricks barely move the answer. Being talked about, credibly and in the right places, is far more closely linked to it. Prisma maps where.
tested optimisation cases significantly improved ranking in a controlled AI-search benchmark.
C-SEO Bench, NeurIPS 2025web mentions versus backlinks, correlated with AI Overview visibility.
Ahrefs, 2025 · 75,000 brandsStart with a Diagnosis to see where you stand. Keep improving with Prisma Studio, which measures every change you make against a baseline.
A decision brief on how AI assistants see your business today, for your buyers' real question.
Your Diagnosis becomes the baseline. Studio keeps watching, tests one true change at a time and shows you what most likely moved the answer.
Scroll the sample brief. Each lens reads the AI's answer from a different angle, so you know what to fix first.
Buyer question: “What's the best sofa for a small apartment in São Paulo, delivered fast?”
Casa Aurora is named and its site is cited, but the answer recommends two competitors first. They publish sofa sizes by room area; Casa Aurora does not.
| Competitor A · “fits rooms from 12 m²” | Recommended |
| Competitor B · “compact line, 1.60 m” | Recommended |
| Casa Aurora · “free 48-hour delivery” | Cited only |
| Free delivery within 48 hours in São Paulo | Supported |
| Its sofas fit most small apartments | Partial |
| Customers praise its service | Unverified |
| It does not offer assembly | Unverified |
Publish a sofa-size guide by room area, a page anyone can verify. Then ask the same question again with Prisma Studio and compare against today.
See exactly where the AI leaves you on the ladder, for the question your buyers really ask.
The brands in the answer and the reasons the AI gives for choosing them.
Every claim checked against the sources it rests on: supported, partial or unverified.
The gap most likely to decide the sale, turned into a concrete, verifiable change.
The answer will change whether you act or not. Studio turns that into a cycle you control, with a receipt at every turn.
The same buyer questions on fixed dates, every original answer kept.
One verifiable change to your site, fingerprinted before and after.
Checks at +7, +14 and +21 days: is the new fact being picked up?
At +28 days, the same design as the baseline, next to your competitors.
A receipt for the cycle, and the next fact to test.
Every number links to the original answer it came from. You can check it without trusting us.
Two AI readers from different providers and one blind human read each answer. Disagreements stay open until settled.
One change at a time, always against a baseline, so you can see what most likely moved the answer, not just that it moved.
The money is pouring into GEO. Public proof of what actually changes an AI's answer is still scarce. That is where Prisma competes.
Tell us what you sell and who you compete with. We come back with how to start your Diagnosis.
