Perception
When someone asks an AI model about your category, it does not only recommend brands — it describes them. Perception shows you that description: what the models associate with your brand, what they hold against it, and whether what they say is true.
For many buyers the AI's description is their first impression of you, and they rarely see the sources behind it. Perception breaks that description into three tabs, each answering a different question:
- Market — what does AI associate with your brand?
- Objections — what does AI argue against your brand?
- Fact-checking — is what AI says about your brand true, measured against facts you state yourself?
All three follow the model filter at the top right of the page, so you can compare how ChatGPT, Gemini, AI Overview or Perplexity each talk about you.
How Perception works
Perception is built from the answers to your tracked prompts — no extra questions are asked. It runs in three steps:
- Read. From every answer that names your brand, Perception extracts the short statements the AI makes about it — "delivers in 24 to 48 hours", "has a large product range", "customer support is slow". These are the claims. A single answer can produce a few of them.
- Group. Different answers say the same thing in different words. Claims that mean the same thing are grouped into one attribute — Fast Delivery, Product Range, Customer Support. Each attribute also carries a polarity: positive, neutral or negative, from the way the AI phrased it.
- Check. If you wrote down your own facts, every claim is compared with them and marked Supported, Contradicted or Not covered.
Perception is refreshed periodically, not after every answer. What you see is the picture as of the last refresh. You do not need to do anything to keep it running. While there is too little data — a new project, or a model with no answers that mention you yet — the page says "Not enough data yet".
Association score
Every attribute gets an association score from 0 to 100. It answers one question: how strongly does AI associate this with your brand?
- A claim raised first in an answer scores 100; each lower position in the answer scores 20 points less.
- The score is averaged over every analysed answer that names your brand — an answer that never raises the attribute counts as 0.
So a score near 100 means the AI brings this up in almost every answer, and near the top. A low score means it comes up rarely or late. The score measures prominence, not truth — that is what Fact-checking is for.
Market
The Market tab is the overview. From top to bottom:
- Headline — "AI describes your brand as Fast Delivery": the attribute the AI reaches for most, and the strongest objection if there is one.
- Most associated — the attribute with the highest association score. This is the top bar in the chart below it.
- Top objection — the negative attribute with the highest score. A dash means the AI raised nothing negative in your tracked answers.
- Attributes — how many attributes exist for the selected model.
- Answers analysed — the answers the scores are averaged over.
- Overall AI Perception — one paragraph that sums up how the models talk about your brand. Green phrases are positives, red phrases negatives.
- How AI describes your brand — one bar per attribute, longest bar first. The badge on a bar is the number of wordings grouped into it; hover the bar to read them.
- Attributes and sources — every attribute as a row. Open a row to see the exact sentences the AI used and the pages the AI cited in the answers where it said this.
If the messaging you invest in does not appear here, the AI has not picked it up yet — and the cited pages are where to start looking.
Objections
The Objections tab shows the negative attributes only: the arguments the AI makes against your brand when a buyer asks.
- Objection mentions — one bar per objection with its association score, computed exactly like the Market scores.
- Sources behind the view — open an objection to see the pages the AI relied on while raising it, with the same columns as the URLs view.
- Terms — the exact words the AI used, the objection each was grouped into, how many chats it appeared in, and on which models. Search it for a specific phrase and sort by occurrences to see which wording comes up most.
An empty Objections tab means the AI made no negative statement about your brand in the tracked answers. Perception only reads the answers to your prompts; it does not go and ask the AI "why not choose this brand?", so the tab stays focused on what buyers actually see.
A false objection is a content and PR problem: find the pages it comes from and fix them. A true objection is product feedback you rarely get any other way.
Sources behind an attribute
When you open an attribute or an objection, the sources table lists the pages the AI cited in the answers where it raised it:
- URL type and Domain type — how the page and its site are classified, as in URLs and Domains.
- Occurrences — how many answers raising this attribute cited the page.
- Retrievals — how many times the page was retrieved across all your tracked answers, not only the ones about this attribute.
- Citation rate — the average number of citations per chat that retrieved the page.
A page with high occurrences is reinforcing that attribute every time the AI reads it — the first place to act if you want to change the picture.
Fact-checking
Fact-checking needs your input. Under Your facts, write the statements you want the AI to get right — one checkable fact per line, up to 50 active: prices, delivery times, return policy, availability, guarantees. Perception then compares every claim with your facts.
- Your brand delivers in 24 to 48 hoursSupportedFact: delivery in 24-48 hours nationwide.
- Your brand delivers within 5 working daysContradictedFact states 24-48 hours, not 5 days.
- Your brand sells refurbished laptopsNot covered
Each claim gets a verdict:
- Supported — a fact confirms the claim.
- Contradicted — a fact directly opposes the claim: a different price, limit, date, availability or feature.
- Not covered — no fact speaks to the claim. Add one if the topic matters.
- Not checked yet — the claim has not been compared with your facts yet.
Three views help you work through them:
- Contradicted — the conflicting claims, newest first, each with your fact and a one-line reason. Start here: pricing, availability and guarantee claims are the most costly to have wrong.
- By fact — each fact with the claims it confirmed or contradicted. A fact that never comes up is either not what the AI talks about, or a topic your prompts do not cover yet.
- By attribute — each attribute with its supported / contradicted / not covered counts.
Open any claim to read the answers it came from. If you disagree with a verdict you can set it yourself; Restore automatic verdict undoes that. Editing a fact re-checks only the claims that fact can change.
Best practices
- Start with the objections and the contradicted claims. They name the exact pages and sentences that cost you buyers.
- Then read what you are not known for. An attribute you invest in but that scores low is messaging the AI has not picked up.
- Check the models one by one. A brand can be perceived very differently by ChatGPT and by Gemini; the "All models" view can hide that.
- Follow attributes to their sources. A low or negative score becomes actionable once you know which pages the AI is reading.