A wrong answer costs you a customer who never calls to check. Eidomai reads what the engines say about you, catches it, and gets the correction sent. You stop wondering what they are telling people.
Every word wrong. A buyer asking which brokerage to use was told this business drives children to school. The same answer went on to describe real-time bus tracking and route optimization.
Every month, every fact the engines state about you is checked against what you told us is true.
See 29 real wrong answers, in the engines’ own words“They do not currently accept Medicare and operate from a single location.”
“Northshore offers aquatic therapy at its Encinitas clinic.”
Sample dataNorthshore Physical Therapy is fictional.
Not whether the answer sounded friendly. Whether it was true. A person at Eidomai reads every one.
Not scored for mood. Read. Each factual statement about you becomes a claim: a service, a location, a name, an insurance panel, a price, an hour. The fact sheet is yours and you confirmed it. Nothing is judged from memory, and nothing you do not publish yourself is ever published by us.
All 48 got a documented fix plan.
A business that changes what it does keeps its old identity in the answers for months. Old profile pages, old press and old listings keep teaching the engines who you were. We check whether AI describes what you do now, and every claim gets one of four verdicts.
Legacy answers cite the 2022 directory listing. Outdated, not wrong.
Sample dataNorthshore Physical Therapy is fictional.
Matches the fact sheet you confirmed.
Never true. Traced to the page that taught it, and corrected there.
True of the old business, not the current one. Usually an old page that needs a redirect or an update.
Nothing on the sheet confirms or rules it out. We ask you before anything is said.
At intake you confirm how a business like yours can be described: the categories that are current, and the ones you left behind. Every answer that names you is placed in one of them from its own words, never from what we already know about you. Entity accuracy is the share of those answers that describe you in a current category. It sits next to visibility, not inside it, because more mentions in the old category is not progress.
The usual cause is a page that still answers for the old business. A retired product page that still resolves, a profile on a directory that never heard about the change, a press piece from before the pivot. Engines cite what they can read. When the old page is still there, the old identity is too. The fix is at that page: a redirect to what replaced it, an updated profile, a newer source that says what you do now.
We find the page that taught the engine the wrong fact, send the correction there, and ask the engine again next month.
Sample data. Northshore Physical Therapy is fictional.
Behind many wrong claims is a page the engine read: a stale directory, an old press mention, a page on your own site that was true in 2021. When the engine cites that page, we send the correction there. When it cites nothing, we say the source is unknown and start with your own site.
A sample from the fictional Northshore Physical Therapy set. Two claims wrong in one answer, both sourced to a 2022 directory listing, both fixable at the source. This is what the Accuracy tab on your portal shows for your business.
Sample data
Not scored for mood. Read. Each factual statement about you becomes a claim: a service, a location, a name, an insurance panel, a price, an hour.
The fact sheet is yours and you confirmed it. Correct, incorrect, or unverifiable. Nothing is judged from memory, and nothing you do not publish yourself is ever published by us.
Engines cite what they read. A wrong claim almost always has a page behind it: a stale directory, an old press mention, a page on your own site that was true in 2021.
A correction asset is drafted for that source, the update is requested, and the claim is re-measured after the next index. The ledger records whether it moved.
29 real wrong answers, quoted word for word, with what is actually true beside each one. Free, no signup.
Every word wrong. A buyer asking which brokerage to use was told this business drives children to school. The same answer went on to describe real-time bus tracking and route optimization.
Not a school in sight. It is a locally owned childcare provider. Parents looking for daycare were pointed at a donation page instead of a tour.
What it is, how it differs from sentiment monitoring, how a correction reaches an engine, and how often claims are checked.
A monthly audit of the factual claims the AI engines make about your business in the answers we sample (your question panel, eight engines, asked by name): services, locations, staff, insurance, hours, prices. Each claim is judged against a fact sheet you confirmed, and every wrong claim gets a correction plan tied to the source the engine cited.
Sentiment tells you whether an answer sounded friendly. Accuracy tells you whether it was true. An engine can warmly and confidently tell a patient you do not take Medicare when you do. That scores as positive sentiment and costs you a patient.
You cannot edit an engine. You can change what it reads. We trace the wrong claim to the page the engine cited, draft the correction for that source (your own site, your Google Business Profile, a directory, an association page), request the update, and re-measure after the next index.
A claim that was true of your business before it changed: a product you retired, an office you left, the category you used to be in. It is not a fact the engine made up, so it is not marked incorrect. It is marked outdated and listed apart, because the fix is usually a redirect or an updated profile at the old page the engine keeps reading.
The share of AI answers that name you and describe what you do now. Each answer is placed in one of your own categories, current or legacy, from its own words. It is scored apart from visibility, because a business can be named more often and understood less in the same month.
Monthly in the audit, with a weekly watchdog on Done for you that catches new claims between audits.
Yes. Every entry is a real "incorrect" verdict from the July 2026 audits, quoted word for word with names removed. 398 claims were checked and 48 came back wrong.
Free check on ChatGPT and Claude, results by email, no call required.
It starts with an eye. It ends with AI. Eidomai, by Nimbletoad.