Reputation

Reputation checks the statements AI answers make about your brand against the facts you have confirmed. When answers repeat a claim, Cleotic groups the repeats into a narrative, so you can see which claims keep coming back, on which engines, and what is true instead. Open Reputation in the brand sidebar.

Reputation is on Business and above. It replaces Perception, and the perception scores are now its Scores tab.

How claims are checked

Cleotic reads every captured answer that names your primary brand. It picks out the statements the answer makes about your brand, quoted exactly as the answer wrote them, and checks each one against your confirmed facts. A statement can match your facts, contradict them, mix your brand up with another company, be out of date, have no fact to check it against, or be an opinion. Statements that name only a competitor are left out.

The checks can only be as good as your facts. A statement with no confirmed fact behind it is marked as having no fact behind it. A negative statement that matches your facts is fair criticism and doesn't become a narrative.

Claim checks use your organisation's monthly model budget. If the budget runs out, the contact check still runs, and the other checks wait until there is budget again.

The contact check

The contact check finds phone numbers and email addresses that an answer presents as your brand's own. It uses fixed rules, not a model.

Only two cases raise a critical alert:

• A phone number presented as yours that isn't one of your confirmed numbers for that country. If you have confirmed a UK support number and an answer gives a different UK number as yours, Cleotic flags it. If you haven't confirmed a number for that country, it doesn't.
• A free email address, such as a Gmail or Outlook address, given as your contact address.

Everything else is unverified, such as a number for a country you have no confirmed number for, an email address on another domain, or a sign-in link. A number given as a warning, or as a regulator's, a branch's or a press office's, isn't presented as yours either. Unverified details are listed on the answer as Contact detail not on file, with Add a fact, so you can confirm the ones that are yours. They never raise an alert.

Add your contact details in Brand settings → Facts → Contact details: phone numbers with their country code, email addresses, sign-in links and domains. Until you have added a support phone, a support email and a sign-in link, Reputation reminds you. Confirming a detail an answer gave closes its narrative.

Narratives

A narrative is one claim repeated across answers and engines, such as a wrong support number or an old price. Narratives is the first tab.

The top line says how many of the answers that named your brand in the period carried a risky claim, with the open risk narratives counted by severity. Below it, Risk narratives lists each one with its severity, its title and a quote, its kind and state, a cell for each engine, its last 12 weeks, and the source the answers cite most.

Each engine cell reads "k of n", meaning k of that engine's latest n answers carry the claim. Cleotic counts the latest 10 answers at most, from the last 28 days, to the questions the narrative appears in. A dash means the engine has no recent answers to those questions.

A risky claim seen in only one answer is held back until it repeats, except a wrong contact detail. The count of held-back claims shows under the table. What to protect lists your strengths: positive claims the engines repeat.

Kinds

Severity

Each risk narrative is Critical, High, Medium or Low. Severity comes from the kind of claim and its topic. A wrong contact detail is always critical, and so is a false or mixed-up claim about whether the company exists, a legal matter or safety.

Severity rises one band for each of these, but never above high:

• Three or more engines carry the claim in at least two answers each.
• At least half the answers that carry it are to buying or trust questions.

States

A resolved narrative that appears in an answer again becomes emerging again.

Read a narrative

Select a narrative to open it. The page shows its severity, kind, state, the day it was first seen, and the count for each engine.

• What the answers say quotes the latest answers from each engine as they were captured. Open answer shows the full answer.
• What's true shows the confirmed fact the claim contradicts. If no fact covers it, link to Facts and add one.
• The cited pages show what Cleotic found when it read them. See Sources.
• Questions it appears in and Alerts list the questions behind it and every alert it raised.

Two buttons record your team's view. It's correct says the claim is true. Not about us says it is about someone else, such as another company with a similar name. Either one closes the narrative's alerts and moves it down the list, and the page shows who chose it and when. Undo reopens the narrative.

Start a correction opens a correction in Opportunities, assigned to you, and records each engine's count at that moment. You can't start one on a narrative marked correct or not about us. Once a correction is open, the narrative page shows Did it land and the Correction request log. See Corrections.

Sources

The Sources tab asks "Does the source say it?" for the pages cited in answers that carry a critical, high or medium claim. Once a day, Cleotic reads the three most-cited pages behind each of those narratives, up to 20 checks a brand each day. It reads each page again at most once a week, and it respects robots.txt.

Source says it and Source says something else show the passage they rest on.

Cleotic doesn't yet judge that a page doesn't say something. It reads the main article text of a page, which leaves out sidebars and comment threads, where many claims about brands are written. It also can't always tell a sign-in wall or a redirect from the page itself. Proving that a page doesn't say something needs the whole visible page, so when Cleotic can't find the claim, the verdict is Couldn't check with the note "Couldn't confirm the page doesn't say it."

Probes

Probes are extra questions Cleotic asks every engine once a week to see what answers say about your brand when people ask about it directly. Their answers feed Reputation. They never count in visibility or against your prompt limit.

The Probes tab offers four packs:

• Trust, such as "Is legit?" and " complaints"
• Support, such as "How do I contact support?" and " login"
• Company, such as "Who owns ?" and "Has had a data breach?"
• Buyer, such as "Is worth the money?" and "What are the downsides of ?"

Ask your own question adds one of yours. Each question must name your brand, can't name a competitor, and is 3 to 300 characters. You can add up to 10 at a time. Stop stops asking a probe.

Business includes 24 probes per brand. Custom plans have no fixed probe limit.

When the brand's site has Google Search Console connected, From Search Console suggests searches for your brand from the last 28 days that contain complaint or trust words, such as refund, scam or login, with their impressions.

Scores

The Scores tab shows your brand's perception scores, which rate how AI models assess the brand on six dimensions, each on a ten-point scale. These are model assessments, not customer survey results.

1. Trust / Reliability
2. Innovation
3. Value for Money
4. Customer Experience
5. Market Leadership
6. Recommendation Likelihood

Scores can include a chart of all six dimensions, a trend over time, AI-written summaries, and the evidence behind a dimension. Choose Week, Month or Year. A missing score or pending summary is not a zero.

Scores refresh on their own schedule, separate from prompt collection. Set it in Brand settings → Reputation, under Sentiment schedule. Business refreshes monthly. Custom refreshes weekly by default and can choose monthly.

Corrections

A correction is an item of kind Correction in Opportunities. It moves through the same stages as other kinds, and its score comes from the narrative's severity. Its play shows the claim, your confirmed fact, and three parts:

• Fix pages. Write a fix page starts a page in Content Studio that opens with the confirmed fact and the claim. A contact correction uses your confirmed contact details. Each fix page shows when it went live and which AI crawlers have read it.
• Did it land. Each engine now, against the day the correction opened: Gone, Fewer, Still says it, Never said it or No recent answers.
• Correction requests. The dated log of every request your team has sent. Log a request opens the narrative page.

On the narrative page, log each request with its channel, who you sent it to, the day you sent it and what you asked. Channels are Engine feedback, Provider form, Source owner, Review platform, Forum reply, Wiki, Letter and Other. Record the reply when it comes: Acknowledged, Changed or Declined, with its date and a note. The first request moves the correction from Now to In flight.

Cleotic never sends correction requests. You send them, and Cleotic keeps the record.

Evidence packs

Select Evidence pack on a risk narrative to build a record of it. Cleotic builds the pack when you open it and keeps no copy. Download JSON saves the data, and Print or save as PDF prints it.

A pack holds:

• A summary, with each engine's counts within the history your plan keeps
• Your confirmed facts, each with its proof: where it came from, the page and passage, and the day it was confirmed
• The cited pages Cleotic read, and what it found on each
• Verbatim captures: up to the 50 newest answers that carry the claim, plus the first one. Each has its UTC time, engine, model, locale, question ID, answer ID, the quoted claim, its citations and a SHA-256 hash of the answer text.
• The correction request log

The pack ends with its disclaimers. It isn't legal advice, and Cleotic makes no judgement about whether a statement is lawful. A capture shows what the engine returned to Cleotic's automated question at the time shown, not what any particular person saw, because answers vary between runs, places and people. A hash shows that the answer text hasn't changed since Cleotic stored it. A source check records what Cleotic could read on a page when it looked, as context for your review, not proof of where an answer came from.

Reputation in reports

A report can include a Reputation section. It is off until you turn on Include reputation in the report's editor. Because anyone with the link can read a report, the section leads with your confirmed facts and never quotes a risky claim.

Alerts

Reputation can email you and send signed webhooks when a risky claim starts repeating. Alert emails and webhooks aren't switched on yet. You can choose your level and add endpoints now, and Send test works. Every alert is still recorded under Alerts on its narrative's page.

Alert emails

Choose your level in Brand settings → Reputation, under Alert emails. The setting is yours alone, for this brand.

A wrong contact detail alerts on its first sighting. A critical claim alerts once it is in two answers from one engine, or on two engines. A high claim alerts once one engine carries it in two answers and it either reaches a second engine or a third answer. Medium and low claims never send an instant email. The weekly email includes a Reputation section.

Your level also covers a narrative that returns after it resolved or rises in severity, and a critical one that reaches another engine. When a critical narrative has gone from the answers, the people who got its first alert hear about it. When a correction lands on an engine, its assignee gets an email.

Webhooks

Add an endpoint in Brand settings → Webhooks with a label, an HTTPS address, and the events to send: Alerts fired (reputation_alert.fired), Alerts resolved (reputation_alert.resolved), or both. Copy the signing secret when it is shown, because it isn't shown again.

Each delivery is a POST with Webhook-Id, Webhook-Timestamp and Webhook-Signature headers. Webhook-Signature is v1, followed by the base64 HMAC-SHA256 of {timestamp}.{body}, keyed with the secret. For 24 hours after you rotate a secret, the header also holds a second signature from the old secret, separated by a space. Accept a delivery when any one of them matches.

The body names the alert, its trigger and severity, the narrative, each engine's counts, a quote, the answer IDs, the confirmed fact, the most-cited source's address, and a link to the narrative. A resolved alert adds why it closed: the claim has gone, your team decided, or you confirmed the contact detail.

Send test sends a signed test event. You can pause, resume, rotate and delete each endpoint, and each one shows its last delivery.

Plans

Reputation, its alerts, webhooks and contact details are on Business and above. On Starter, Reputation shows as locked. If a brand moves to Starter, Reputation locks and new checks stop.

Connected agents can read Reputation through get_reputation and get_narrative. See the MCP tool reference.

What Cleotic can't do

Cleotic can't change what an AI engine says. Each engine decides its own answers. A correction helps you publish what is true and keep a record of the requests you send, and Did it land shows whether the answers changed afterwards.

Cleotic sees only the answers to its own questions. A narrative shows what the engines told Cleotic, not what every person who asked was told.

Related

• Facts
• Answers
• Opportunities
• Reports
• Brand settings