Private beta · now open
VritantAI Discover
Where you stand in AI answers, why, and what to fix first.
Discover asks the AI engines on your plan (ChatGPT, Gemini, Perplexity and Claude, depending on the plan) the questions your customers ask, more than once a week, then shows how often each one mentions you, links to you and recommends you. Below: each part of an account, how the numbers are made, what is counted and what is estimated, and how your data is handled.
- ChatGPT
- Gemini
- Perplexity
- Claude
- Request-based beta access
- Discover never edits your site
- No ranking or citation guarantees
Module by module
What each part of an account does.
- Beta
Measure every engine, many times
Every engine on your plan answers each of your questions more than once a week. You see how often they mention you, link to you and recommend you, each with a 95% range, and a verdict on each change for every engine and for your business overall: real change, possible change, or within normal noise. A single question often stays preliminary until it has enough answers.
- You pick the questions: ready-made templates for online stores, brands and local businesses, or your own, within your plan's limits.
- Every answer is stored with the engine, the date and how it was collected, so you can open the exact text behind any number.
- Each answer is checked for three things: whether it names you, whether it links to your site, and whether it recommends you.
- Rates are shown per question, per engine and for your business overall, each with a 95% range and a change verdict.
- Each engine's search rate shows how often it searched the web before answering, because an answer from memory cannot cite a page you just updated.
- Beta
Catch wrong facts
Each week, for every question and engine, Discover checks an answer that mentions you against the facts you confirm: hours, locations, prices and policies. Each claim is marked correct, wrong or unverifiable, next to the answer that made it.
- You confirm the facts AI should get right: opening hours, locations, phone numbers, prices, offerings and policies.
- Each claim an answer makes about those facts is marked correct, wrong or unverifiable, with your confirmed value beside it.
- The share of correct claims carries a 95% range too, so one bad answer does not read as a trend.
- A wrong claim raises an alert and becomes a diagnosis finding with a fix.
- Beta
See why, then fix what matters
A six-stage diagnosis for each engine, from whether AI crawlers can reach your site to whether answers link to you. Most findings come with a fix, its steps and an evidence label, and the three most useful fixes come first.
- Each stage gets OK, Weak, Issue or Unknown for every engine. Unknown means Discover has no data for that stage yet, and it says what data is missing or what to connect.
- Every finding links to its evidence (the access check, the answer or the question) and says whether you can fix it, can influence it, or it is outside your control.
- Fixes are ranked by impact, strength of evidence and effort, with steps and, for some, a template: a robots.txt snippet, a listing update checklist, a review request message.
- You move a fix through To do, In progress and Done. When the next diagnosis finds the problem resolved, Discover records the number the fix should move and marks it Measuring. If the problem returns, the fix goes back to To do.
The stages
- AccessibleCan AI crawlers read your site?
- IndexedAre your pages in the search indexes engines use?
- MatchedDo your pages answer the questions people ask?
- SelectedDo engines pick you over other sources?
- RepresentedDo answers describe you correctly?
- ConvertedDo answers send people to you?
Discover Growth includes AI drafts for content fixes, drafted from the facts you confirm, with anything else marked for you to check, and never published for you.Beta
- Beta
Competitors and the sites AI cites
Share of voice against the competitors you track, with ranges, plus competitors the answers keep naming that you have not added. The citation gap lists the third-party sites AI cites most for your questions and, for the top sources, whether you appear on the page of each that AI cites most.
- Share of voice against the competitors you track, each with a range and a verdict, plus each one's recommendation rate.
- Suggested competitors are the names answers keep mentioning that you have not added. You accept or dismiss each one.
- A competitor's numbers start when you start tracking it: earlier answers are not read again.
- A site that does not name you on the page AI cites most is reported as missing from that page, not from the whole site.
- Beta
A monthly report you can share
Each closed month becomes a fixed report with ranges, verdicts, wrong facts and fix progress. Download it as a PDF or send a share link that expires, with an optional password. A weekly digest arrives by email, with alerts when a drop is real, an answer states a wrong fact, or an AI crawler is blocked.
- Each closed month becomes a fixed snapshot that never changes after it is published.
- It covers your headline rate against the previous month, each engine, the biggest real changes by question, share of voice, wrong facts, crawler access and fix progress, plus what the month's data does not show.
- Download it as a PDF, or send a share link that expires, with an optional password. You can revoke a link at any time.
- Email alerts go out for a real drop, a wrong fact, a blocked AI crawler, or a Google connection that needs reconnecting. A competitor overtaking you and newly suggested competitors appear in the app and the weekly digest.
- Beta
AI traffic from your Google tools
Connect Search Console and GA4 with read-only access to see visits that AI assistants referred and your branded and non-branded search clicks next to your mention rate. They sit side by side as correlation, never presented as cause and effect. Until Google verifies our connection, it is open to invited beta accounts only.
- Read-only connections to Search Console and GA4. Discover never writes to your Google accounts.
- Visits referred by AI assistants come from GA4, by source and as a share of all visits.
- Branded and non-branded search clicks come from Search Console. Clicks on queries Google hides, and on queries beyond the rows Discover reads, are shown as unclassified, never guessed.
- Each week shows how many of its days are synced, so a partly synced week is never read as a drop.
How measurement works
Repeated samples, ranges, and a verdict on every change.
AI engines can give a different answer to the same question minutes apart, so one answer is a single sample. Answers are collected through each engine's official API with web search turned on.
Sample
Each question goes to every engine on your plan more than once a week, spread across the week, and every answer is stored.
Count
Each stored answer is read for a mention of you, a link to your site and a recommendation. Each week, one answer per question and engine that mentions you is also checked against your confirmed facts.
Estimate
The counts become rates with a 95% range: the true rate is likely inside it. More answers make the range narrower.
Judge the change
Every week, the last two weeks are compared with the two weeks before, per question, per engine and for your business overall, and each change gets a verdict.
- Real change
- The 95% range of the change excludes zero, and the change is large enough to detect with this many answers.
- Possible change
- The 80% range excludes zero, but the change does not meet the bar for real. The data leans one way but could still be noise, so check again next week.
- Within normal noise
- The difference fits within normal week-to-week variation.
- Preliminary
- Not enough complete, comparable data yet to judge a change.
A change stays preliminary when any of these is true:
- fewer than 10 answers in the period
- fewer than 90% of the scheduled answers arrived
- answers from fewer than 3 different days
- no answers yet in the period it is compared with
- a different collection method, engine version, location or question wording than the period it is compared with
- an engine change was detected in either period
Mention rate by engine, last two weeks
- ChatGPTWithin normal noise
22% of answers, likely between 14% and 31%
- GeminiReal change
41% of answers, likely between 32% and 50%
- PerplexityPossible change
35% of answers, likely between 26% and 45%
- ClaudePreliminary
9% of answers, likely between 3% and 19%
Counted versus estimated
Which numbers are counts, and which are estimates.
Mentions, links and wrong facts are counted from stored answers you can open. Rates are estimates from those samples. AI referrals from GA4 are a floor: visits from AI apps that send no referrer appear as direct traffic.
- Counted
- Answers collected, mentions, links, recommendations, wrong claims
- Exact counts from stored answers you can open and read.
- Estimated
- Mention, link and recommendation rates, share of voice, share of correct claims
- Estimates from those samples, each with a 95% range.
- A floor
- Visits referred by AI assistants, from GA4
- The real number is at least this. AI apps that send no referrer show up as direct traffic.
- Correlation
- Search clicks and AI visits next to your mention rate
- Shown side by side. Moving together does not prove one caused the other.
Evidence labels on every fix
How sure we are that a fix helps.
Every fix names its evidence label and the source behind it, so you can decide what is worth your time. Fixes with stronger evidence rank higher.
- Tested
- A controlled study or the engines' own documentation shows the effect.
- Observed
- Linked with better visibility in observed data. An association, not a proven cause.
- Unproven
- Plausible, with no published evidence yet. Never ranked in the top three unless it takes minutes.
- Tested, no effect
- Tested and found not to help AI visibility. Listed under “Don't bother” so you skip it.
Your data
What Discover reads, keeps and deletes.
SOC 2 Type II in progress · Data encrypted at rest · Tenant-isolated by design · No training on your data
Security and data handling- Discover never edits your site
- It reads public pages and AI answers. Every change is yours to make: fixes come with steps, and AI drafts are saved for your review, never published.
- Google connections are read-only
- Search Console and GA4 are connected with read-only access. Disconnecting deletes the data we synced from them.
- Share links are yours to end
- A report link expires on the date you set and can be revoked at any time. A password is optional.
- Deleting your workspace deletes your answers
- Deleting your workspace removes your stored answers and their archives once its 30-day cancellation window ends.
Not available yet
These are built or planned but not part of a Discover account today. Some are built and switched off while they are tested. The list is read from the same registry as the badges above, so it changes when they ship.
- Fix Lab before-and-after experimentsRoadmap
- Shared question panels by city and categoryRoadmap
- Agency client workspaces and white-label reportsRoadmap
- Stale-answer detectionRoadmap
- On-site "how did you hear about us" widgetRoadmap
- Google Business Profile importRoadmap
- Applying approved fixes to your site from DiscoverRoadmap
Free, without an account
Start with the free check.
Free check of whether AI assistants name your business when customers ask about your category. We ask several AI engines a set of real customer questions, twice each, and show how many answers named you, a competitor and which sites they cited, as plain counts.
Free AI hallucination scanner: Reads one product page, runs a single-engine AI check, and flags claims the page does not support.
Pricing
Plans sized by questions, locations and engines.
Founding-cohort pricing during private beta. Request access; listed rates are not public checkout offers.
Trial
See how AI answers describe your business, across every engine we measure.
$3one-time
Does not renew
₹299 one-time in INR · Does not renew, plus 18% GST
- 1 business
- 1 location
- Up to 5 tracked questions
- Up to 2 competitors
- 4 AI engines: ChatGPT, Gemini, Perplexity, Claude
- Citation gap: top 10 third-party domains
7 days of access for a one-time payment. It does not renew. A trial shows a first baseline; confirming a change over time needs at least two measurement windows, which a trial is too short to show.
Request beta accessStarter
Weekly AI visibility measurement for one business and one location.
$39/mo
Billed monthly
₹1,499 /mo in INR · Billed monthly, plus 18% GST
- 1 business
- 1 location
- Up to 10 tracked questions
- Up to 3 competitors
- 4 AI engines: ChatGPT, Gemini, Perplexity, Claude
- Citation gap: top 10 third-party domains
- Billed in INR: 2 AI engines: ChatGPT, Gemini
Growth
Breadth across more questions and locations, with AI drafts for fixes.
$149/mo
Billed monthly
₹4,999 /mo in INR · Billed monthly, plus 18% GST
- Everything in Starter, plus
- 3 locations
- Up to 50 tracked questions
- Up to 10 competitors
- Citation gap: top 25 third-party domains
- AI drafts for content fixes
Agency
Client workspaces and white-label reports are not built yet. Talk to us about what you need.
Talk to usNo public price. Tell us what you need.
Discover is priced in USD and INR. Your billing currency is set at checkout from your billing country. INR prices exclude 18% GST. Discover Starter and Growth plans billed in INR measure ChatGPT and Gemini. There is no overage charge: when a plan's monthly measurement budget is used up, sampling pauses until the next month and the dashboard shows the gap. Compare plans in detail
FAQ
Questions about Discover.
How often does Discover ask each question?
Every engine on your plan answers each tracked question more than once a week, spread across the week, and every answer is stored. How many questions, locations and engines you track depends on your plan.
Which engines does Discover measure, and how?
ChatGPT, Gemini, Perplexity and Claude, depending on the plan. Answers are collected through each engine's official API with web search turned on. An engine's own app can answer a signed-in person differently, so Discover reports a sample of what each engine says, not what every customer sees. Google AI Overviews and AI Mode are not measured yet.
When does Discover call a change real?
When the 95% range of the change excludes zero and the change is large enough to detect with that many answers. When only the 80% range excludes zero, or the change is too small to count as real, it is a possible change, and otherwise it is within normal noise. With too few or incomparable answers, the change stays preliminary.
What counts as a wrong fact?
A claim in an answer that mentions you and contradicts a fact you confirmed, such as your opening hours, a price or a policy. A claim Discover cannot check against a confirmed fact is marked unverifiable, not wrong.
Why does a fix say Observed or Unproven?
Every fix carries an evidence label with its source. Tested means a controlled study or the engines' own documentation shows the effect; Observed means an association, not a proven cause; Unproven means plausible without published evidence yet. Tactics tested and found not to help are listed under “Don't bother”.
Can I use Discover now?
Discover is in private beta. We accept a small number of businesses at a time, and each one works directly with the team. Request access below. The free AI visibility check needs no account.
How much does Discover cost?
Discover plans start at $39/month, and a 7-day trial costs $3. Founding-cohort pricing during private beta. Request access; listed rates are not public checkout offers.
Private beta · now open
Want Discover for your business?
We accept a small number of businesses at a time, so each one works directly with the team. Tell us your website and we will reply with next steps.
- Request-based beta access
- Discover never edits your site
- No ranking or citation guarantees
Learn the ground rules
Guides to AI visibility.
- What is generative engine optimisation (GEO)?
- What a GEO platform does for an e-commerce store
- A step-by-step Perplexity guide for e-commerce brands
- How to track whether ChatGPT mentions your brand
- Detecting and fixing fabricated product claims
- Schema markup and JSON-LD for product pages
- Copy-paste JSON-LD for Shopify and WooCommerce
- llms.txt for e-commerce: should you bother?
- GEO tools for e-commerce compared