A/B testing for your AI funnel

The funnel moved off your site. See it. Then test what changes it.

People ask AI to compare and decide before they reach you. Oomira shows what it said, why, and whether your change moved it.

Your first question is free. No account. Two minutes. Then 7 days of Teams free, no card.

A/B test · one question, 4 agents, asked cold“What does this company do?”
ABefore

“Pivoting. Positioning is unclear.”

What the 4 agents said. Your site had already changed; the pages machines read first had not.

what machines read firstOne change
meta description
oldcurrent
agent card
oldcurrent
BAfter · asked again, cold

“They describe the current product.”

Same 4 agents, same question, minutes later. Every one opened your pages fresh.

Hypothesis: machines were reading an older site than people see. One change, then the same question, asked again.

Result · supported
Asked onChatGPTClaudeGeminiPerplexity
How it happens

The comparison now happens inside the answer, before anyone reaches your site.

A person, on your contentAI, reading for themAI agent, soon
  1. A buyer asks“Which invoicing tool should a 20-person agency use?”
  2. The agent readsYour pages 3 times. Outside pages 22 times.
  3. The agent decides“Use the other one: their API is read-only.” 3 of 4 agents.
  4. The buyerNever visits. Nothing in your analytics.
Before AI
DiscoverConsiderDecideBuy
Seen inSearch consoleYour siteYour siteYour site
Now
DiscoverConsiderDecideBuy
Seen inAEO toolsOomiraOomiraYour site
Across the 20 questions

4 against you, 11 mixed, 5 for you. Every against verdict is a buying decision: the comparison, the API, security, fees.

What follows

The answer is an output of specific pages, yours and other people’s. So it can be changed, and the change can be measured.

Where this fits

Analytics measures the visit. AEO measures visibility. Oomira is A/B testing for AI answers.

ToolThe question it answersThe unit it works in
Web analyticsWhat people do once they reach your siteVisits, sessions, conversions
SEO and technical toolsWhether your pages are indexed, reachable and structuredPages and rankings
AEO and visibility toolsWhether AI mentions you, and your share against competitorsMentions and share, over time
OomiraDoes changing this page change what the agent says? Change it, re-ask, compare.Experiments: a change, a prediction, a result

They work together: Oomira finds the page and runs the test, and the tools you already have show the traffic move.

Finding 1 · Who supplies the answer

151 of the 212 pages agents opened were not yours. The affiliate list they use is not the one you pay.

212pages opened
151not yours
12in your affiliate program
3of those ever quoted
9unpaid, quoting you well
2quoting a wrong claim
review roundupcomparison sitesoftware directoryforum threadtech magazinea 2024 comparisonlaunch communityfounder blogencyclopedia entrynewsletterprofessional networkratings sitecode hostcompany databasedev communityemployer reviews
Helps youHurts youNeutralIn your affiliate programSize = how often agents opened it
PagePaidOpenedQuotedEffect
A review roundup, this yearNo4/44/4Helps
A comparison site, 2024No3/43/4Wrong API claim
The affiliate angle

You pay 12 partners and agents quoted 3 of them. A review roundup, a tech magazine and a founder blog, none of them paid, shaped 5 to 8 answers each. One 2024 comparison carries 6 of the against-you answers on its own.

What this gives the affiliate and PR team

A ranked list of the sites that decide the answer, what each one says about you, and which of them are wrong.

Finding 2 · Where answers turn against you

Four buying decisions go against you. Each one traces to a page.

You3
Comparisons won
of 11 head-to-head questions
8One competitor
ChatGPT · to an agency owner“Which invoicing tool should a 20-person agency use?”

They are the fastest to set up, but their API is read-only, so most agencies end up elsewhere.

What it rests on
A comparison site, 2024not yours · quoted 3 of 4
your /onboardingyour page · quoted 4 of 4
your /changelogwrite API, May · opened by 0 of 4
3 of 4 againstA competitor, for buyers who need an API

Settled from a 2024 comparison site. Your changelog says the write API shipped in May. No agent opened it.

4 of 4 againstA competitor, for IT buyers

They list SOC 2, SSO and SCIM on one page. Your /security has SOC 2 in an image.

2 of 4 againstFees for a small team

Two agents quote a per-invoice fee from a 2024 help article that /pricing replaced.

Leads with criticsReaction to the payments launch

A forum thread leads. Your response exists as a PDF that no page links to.

The pattern

In every case the accurate version exists somewhere on your site. It is on a page agents do not open, or worded so they do not use it.

Finding 3 · Your site, as a machine reads it

Most of the site reads fine. What fails is concentrated in the pages agents actually use.

Works with 200+ toolsSlack, HubSpot, Salesforce, Notion and more.See all integrations
$ oomira fetch yoursite.com/integrations
fetch    200 · 0.6s
body     41 words
list     rendered by script · 0 names in HTML
schema   none
opened   2 of 4 agents
quoted   never
1 in 3citations of your pages is a help article

/pricing cited 9 times, the 2024 fee article 7, the API reference 5. Buyers hear your prices in support wording.

19 of 71pages agents opened returned almost no text

/integrations is rendered by script: 41 words and no tool names in the HTML. Agents open it and quote none of it.

6opened pages have moved, and 2 addresses are gone

592 of 624 pages load. The comparison page an agent wants for the head-to-head question redirects to the homepage.

Every item comes with the page address, what the fetch returned, and which agent hit it. Engineering gets a ticket, not a theory.

See your site as a machine
A/B testing for the AI funnel · 4 high-priority tests

One change, one prediction, then the same questions asked again.

What agents say nowThe changePredictionRe-askStatus
Your API is read-onlyState the write API in the first paragraph of /apiFewer answers call it read-only4 questionsSupported
A competitor is safer for IT buyersPublish SOC 2, SSO and SCIM as text on /securityAgents quote your page, not only theirs3 questionsRe-testing
Fees are charged per invoiceAlign the 2024 help article with /pricingAgents stop quoting the old fee2 questionsShipped
You never responded to the launch criticismPublish the response as a page, linked from the launch postFewer answers lead with the forum thread3 questionsOpen

AI answers vary, so a prediction says fewer agents will say it, never that one will. Each test ends supported, rejected or inconclusive against the answers before it.

Who does what

Every finding lands on a team as work, with the evidence attached.

ContentPut the fact on the page agents open

The write API on /api. Current fees in the help article. Security controls as text.

Affiliate and PRWork the sites that decide the answer

Ask the 2024 comparison site to correct it. Brief the review roundup that shaped the most answers. Put the launch response where agents read.

EngineeringFix what agents hit

Render /integrations in HTML. Redirect the 6 moved pages and the 2 dead ones. Restore your own comparison page.

After every changeRe-ask and compare

Same questions, same agents. Answers, pages and sources compared with before, so you see which change moved what.

Pricing

Your first question free. Then 7 days of Teams free.

  1. The answerIts claims marked for you, mixed or against you
  2. The pages behind itEvery page it opened, yours and not, fetched and kept
  3. What it meansFindings, each resting on the page that caused it
  4. What to tryTests with a change, a prediction and the questions to re-ask
Free
$0

Your first question, put to every model.

  • 4asks: one question, every model
  • 4models: Claude, ChatGPT, Perplexity, Gemini
  • The answers annotated: sentiment sentence by sentence, every search run and every page cited
  • Every page they opened, shown.
Pro
$99/mo

One company. Ask, change, re-ask.

  • 160asks a month, on the models you choose
  • 4models: Claude, ChatGPT, Perplexity, Gemini
  • 1domain: all its pages, and every page of it an agent opens
  • 1user
  • Every answer annotated: sentiment sentence by sentence, the searches it ran, the pages it cited
  • Beliefs, conclusions and tests you can run, each resting on the answers behind it
  • Re-ask any time: after a change, or on a new question
  • Unlimited site checks, all fixes, exports
Teams7 days free · no card
$259/mo

For the people who own the pages.

  • 320asks a month, on the models you choose
  • 4models: Claude, ChatGPT, Perplexity, Gemini
  • 1domain: all its pages, and every page of it an agent opens
  • 5users
  • Every answer annotated: sentiment sentence by sentence, the searches it ran, the pages it cited
  • Beliefs, conclusions and tests you can run, each resting on the answers behind it
  • Re-ask any time: after a change, or on a new question
  • Unlimited site checks, all fixes, exports
Custom
Enterprise

Several brands, markets or a custom setup.

  • Everything in Teams, at your volume
  • Every page an agent opens, read and fact-checked against your company record
  • Answers traced past their citations
  • Site checks across your subdomains
  • Owners on each fix, and a re-ask once it ships
  • Compared against competitors

One ask is one question put to one model. A question on all 4 models is 4 asks; a test re-asks only on the models it needs. Spend them when you want, to re-test a change or add a question.

Questions

What people ask before they start.

What does Oomira do?

The customer journey now happens before the visit: people ask an AI agent to compare, recommend and decide. Oomira asks 4 named agents (Claude, ChatGPT, Perplexity and Gemini) the questions buyers, investors, candidates, partners and press actually ask about a company, cold, with live web search and nothing supplied by us. It shows each answer word for word, marks its claims for you, mixed or against you, and traces the answer to the pages it opened: the company’s own, its competitors’ and the outside sites that shaped it. Then it writes a hypothesis about what is driving the answer, and when a page changes it asks the same questions again and compares with the answers before.

How is Oomira different from AEO or AI visibility tools?

Web analytics measures what people do once they reach a site. SEO and technical tools measure whether pages are indexed, reachable and structured. AEO and visibility tools measure whether a company appears in AI answers and its share against competitors. Oomira answers a different question: does changing this page change what the agent says. It works in experiments, a change then a prediction then a result, which is closer to A/B testing for the AI funnel than to a visibility dashboard. They are complementary: Oomira finds the page and runs the test, and the other tools show the traffic move.

What comes back from one ask?

The answer word for word, annotated: each sentence marked for you, mixed or against you, with the searches that agent ran and the pages it cited. Every page it opened is fetched and kept, yours and other people’s, so a claim can be traced to the page that caused it. From the answers Oomira writes beliefs the agents hold, conclusions about the account, and tests to run: each one a change, a prediction, and the questions to ask again.

How is this different from asking ChatGPT about my own company?

Your own assistant holds your history, your earlier questions and often your own website already in the conversation, so it answers about a company it has been introduced to. Oomira asks four agents cold: a blank slate, nothing supplied by us, one live search each. That is the answer a stranger gets, and it comes back with every page each agent opened, so a claim can be traced to the page that caused it rather than argued with.

Can you actually change what AI says about a company?

Yes, by changing what the agents read, and you can see it happen. Every ask is an evidence environment: each search the agent ran, each page it opened and each page it cited is recorded, and each sentence of the answer is traced back to one of them. So a claim is either backed by something that agent just read, in which case the page that caused it is named and can be changed, or it is backed by none of them, which means it came from the model or from something cached, and that is a different problem with a different fix. Change the page or the outside source, ask the same questions again, and compare with the answers before. Answers vary between days, so a test predicts that fewer agents will say the thing rather than that one agent will say a given sentence.

How long does the first read take?

A couple of minutes. Your first question is free and needs no account: 4 agents answer while you watch. The first read of a paid account is 20 questions we generate for your company, which you can edit or replace with your own, each put to every model: 80 asks. After that nothing runs on a clock. Asks and tests run when you press, and new insights keep arriving as those answers land.

If I change a page, how fast does the answer move?

Sometimes the very next ask, sometimes not. An agent that opens the changed page can pick it up immediately; a claim that came from an outside site, or from something cached, moves when that source moves. That is why a change is a test rather than a fix: you re-ask the same questions on the same models and compare with the answers before, instead of waiting and hoping.

What does it cost?

Paid plans are counted in asks. One ask is one question put to one model, so a question answered by all four models is four asks; a test re-asks only on the models it needs. A monthly plan is $99 for one company, with 160 asks a month: the setup questions put to every model, then the asks spent when you change a page and ask the same questions again. A shared plan for a team is $259, with 320 asks a month. The team plan starts with 7 days free and no card.

Why I built this

I used to obsess over every page in the funnel. Then the funnel moved somewhere analytics can’t see.

At Sheertex, I lived in the dashboards. We grew to tens of millions in revenue and raised hundreds of millions, and I constantly changed the pages that shaped what people believed about us: the homepage, product pages, checkout, investor decks, careers pages, press materials.

If a buyer, investor, candidate or partner was forming an opinion of the company, I wanted to know what they saw and what moved them.

Today, a lot of that happens before anyone reaches your site.

A buyer asks ChatGPT or Claude what to use. The agent reads your pages, competitor pages and outside sources, then hands back a recommendation. Nothing appears in your analytics. Soon the buyer may never visit at all: their agent will do it for them.

I originally built Oomira because I wanted my own AI agent to get the facts about me and my company right. Then I realised the bigger problem was everyone else’s agents getting you wrong, in front of your buyers, with no way to see why or test a fix.

Oomira is the same obsession with the funnel, pointed at where the funnel is now.

  1. 01See what agents say.
  2. 02Trace what shaped it.
  3. 03Change one thing.
  4. 04Ask again.
Katherine HomuthKatherine Homuth, founderPreviously founded Sheertex, Female Funders and ShopLocket.

See what AI told your last buyer.

Your first question is free. No account. Two minutes. Then 7 days of Teams free, no card.

Oomira: The funnel moved off your site. See it. Then test what changes it.