What do you think of ?
Half the companies we have measured score 0 or below.
A score is promoters minus detractors over the answers scored, so 0 means about as many agents talk a company down as talk it up. Nobody in either half knew.
The shaded span is the middle half of 199 companies scored so far. The line is the median.
Every page each agent opened, and the ones it could not.
The retrieval log sits beside every answer, and the checks say which of your pages an agent can reach at the moment it answers. A finding is the gap between what is true and what is said.
Each agent chose its own searches. The site checks run separately, from a catalogue of 45.
Ship a fix. Get a verdict.
A trend needs two readings of the same question, so a fix is measured rather than asserted. Nothing is claimed from one reading.
- FindA findingSomething the answers keep getting wrong, or keep leaving out.
- ShipDatedThe date is where the measurement is cut.
- AskThe same questionAsked again afterwards, cold, of the same four agents.
- ReadA verdictJudged against a rule you set before shipping, so it cannot be argued after.
The same sentence, to four agents, judged the same way.
What do you think of yourcompany.com? Answer in 100 words.Sent verbatim. No system prompt, no memory, nothing from us.
Each answer is scored 0 to 10 by a separate model against a rubric published in full, and the clause that set the score is quoted beside it.
The reading is promoters minus detractors over the answers scored, from −100 to +100 — and a question with its own positions is read on those instead.
The report adds your record, the findings, and the site checks that apply to you, from a catalogue of 45.
The things you are already thinking.
Can I not just ask ChatGPT myself?
That is a different question. Your assistant holds your history, your earlier questions and often your own site already in the conversation, so it answers about a company it has been introduced to. These four are asked cold: blank slate, nothing supplied by us, one live search each. That is the answer a stranger gets, and a stranger is who decides.
Agents say something different every time. What is a score worth?
The variance is the finding, not a flaw in the measurement. Four agents are asked in parallel and the spread between them is printed rather than averaged away. Nothing is claimed from one reading, and a change only counts as a fix when the same question, asked again after it shipped, moves.
Can you change what the agents say?
Nobody can, and anyone promising to is selling something else. What is changeable is what an agent can reach at the moment it answers. The report names which of your pages each one opened, which it could not, and where what got said differs from what is true.
Why should I believe the score?
Because it is checkable line by line. The question goes out verbatim and is printed on this page. The rubric is published in full. Every answer is quoted whole, and the clause that set each score is quoted beside it — so you can disagree with a score and see exactly which sentence produced it.
How long before I know anything?
About two minutes, free, with no account: four agents answer on this page while you watch. The full report — your record researched independently, the site checks and the findings — takes about 15 minutes and arrives by email.
Not just what they said. What to change.
Your record researched independently, every answer in full, and the site checks that apply to you — then the specific changes that close each finding, in the order that matters, each with the evidence that says so. Written from the reading you just watched land, and ready to hand to whoever has to do the work.
Read one, whole and unlocked
Start with a report. Keep the important questions monitored. Scale it across the company.
See how AI represents your company, understand what is shaping the answers, and track whether that changes over time.
A complete AI reputation report for one company, one time.
- What ChatGPT, Claude, Gemini and Perplexity say about you
- What they consistently praise, misunderstand or hesitate over
- What public evidence is shaping those answers
- How you compare with relevant competitors
- The most important changes you can make
- Full answers and source trails
Delivered in about 15 minutes.
For one company that wants to keep the important questions under watch — and investigate anything else as it comes up.
- Up to 10 tracked questions measured over time
- Re-run tracked questions whenever you make a change
- See when important beliefs or answers shift
- Ask ad hoc questions anytime without adding them to monitoring
For founders, comms leads and teams actively managing one company.
Tracked questions are the questions Oomira keeps measuring so you can see change over time. Ad hoc questions are anything you want to investigate right now.
For companies with multiple products, brands, markets, teams, and stakeholder questions.
- Multiple products, brands, business units, and URLs
- Separate question sets for investors, customers, candidates, partners, trust, and other audiences
- Company-wide findings that show when the same reputation issue is affecting multiple teams, products, or audiences
- Executive briefings on what changed, what improved, and what needs attention
For companies treating AI reputation as an ongoing corporate function.
Talk to usThe reading on any company, on this browser, with no account: four agents asked cold, every answer scored, the open answers on screen.