Perception

Ask ChatGPT "what do you think of Nike?" and you get a paragraph. Premium, performance-led, expensive, iconic swoosh, mixed on labor practices. Ask it again next week, ask it from a different IP, ask Claude and Gemini and Perplexity the same thing, and you get hundreds of paragraphs that mostly say the same things in slightly different ways.

Perception is what happens when you treat those paragraphs as data. We ask the four major AI models to describe your brand and your competitors side by side, every week, and we score the answers across twenty dimensions: how trustworthy, how innovative, how premium, how easy to use, how recommendable, how unique. The output is a map of how AI talks about you: quantitative, comparable to last week, comparable to the brand you're chasing.

The word "perception" sounds soft. The thing itself isn't. Every score on this page is grounded in a specific AI response, an extracted descriptor, a named concern. When your "Value for Money" drops eight points, somebody, usually a model that scraped a price-complaint Reddit thread, said something specific that pulled the number down. You can read it.

What perception measures

Twenty metrics, grouped into five categories. Each metric is a 0-100 score that says how AI describes you on that dimension, relative to the competitors you're tracked against.

Each category has four sub-metrics. Trust splits into overall trust, reliability, transparency, and safety. Innovation splits into forward-thinking, adaptability, likability, and confidence-inspiring. The full twenty are visible on the Perception page, and each one carries its own score, rank, and trend.

Above the dimensions sits a single Perception Score, 0-100, weighted across all twenty. That's the headline number. It moves slowly; perception is a stock, not a flow, and a five-point swing in a week is meaningful.

How we collect it

Once a week, the perception pipeline asks four AI models, GPT-4o, Claude Sonnet, Gemini Flash, and Perplexity Sonar, to analyze you and your tracked competitors in a single batch. Same prompt, all four models, structured output.

The prompt asks each model to do four things for every brand:

1. Score the twenty metrics from 0 to 100, with explicit guidance to differentiate brands (no flat 70s across the board).
2. Name three key strengths the brand has versus its competitors.
3. Name two or three key concerns or areas for improvement.
4. Give five descriptive words and a one-sentence positioning statement.

Those four streams of output are what every part of the Perception page is built from. The scores become category scores, rankings, radar axes, and trends. The descriptors become Semantic DNA. The concerns become the common-concerns panel. The positioning statements get synthesized into the brand narrative.

Two things worth knowing about the sampling:

• It's a comparative analysis, not a poll. Models are told to score you relative to the specific competitors in the batch. That's why the same brand can have different scores in two workspaces tracking different competitor sets, and why your competitor list materially affects what you see. Track the brands you actually compete with.
• It runs weekly, not daily. Perception moves slowly. A weekly cadence catches real shifts without drowning you in noise from one model's bad day. The exact day depends on when the brand was set up; the date stamp on the page tells you when the current snapshot landed.

What the page answers

Five tabs, each one answering a different question.

Overview: how am I being perceived right now?

The default view. A headline Perception Score, your rank against tracked competitors, and the seven-day change. Below that, the five category cards. Below that, a radar chart that overlays you against the average competitor across all five categories; the shape tells you, in one glance, where you over-index and where you trail.

A trend chart at the bottom plots your score over time, with optional competitor overlays. The model grid on the right shows how each of the four AI models scored you individually. They almost never agree perfectly. Gemini being 10 points kinder than Perplexity is normal and itself diagnostic; Perplexity reads the live web more aggressively, so it picks up bad recent press faster.

Story: what is AI actually saying?

The Story tab is where the words live. At the top, a synthesized narrative: one paragraph that reads like a brand brief written by AI. Below it, your Semantic DNA: the descriptors that show up most consistently across model responses. Words that are growing in frequency are marked emerging; words that are fading get marked too. This is the early-warning system for narrative drift.

Then use case mapping: which use cases AI consistently associates with you, and which it consistently hands to competitors. A common concerns panel lists the specific complaints AI raises about you (expensive, steep learning curve, poor support). A source influence footer connects perception back to the domains that drive it, so you can see which Reddit thread or G2 review is doing the work behind your numbers.

Competitors: how do I stack up, head to head?

A score matrix with every dimension as a row and every brand as a column. Use it to find the specific dimensions where a competitor leads and the gaps that explain why they win in a sales cycle. Head-to-head view lets you pick one rival, see your wins versus theirs, the biggest lead either side has, and a radar overlay.

This is different from the Competitors page, which measures share-of-voice: who gets mentioned more often. Competitive perception measures share-of-favor: when AI does mention you both, who does it describe better.

Goals: what am I trying to move, and am I getting there?

Pick one or more dimensions you want to improve (say, Innovation from 62 to 75), set a target date, and the Goals tab tracks your trajectory. Each goal gets a status: on track, at risk, behind, or achieved. It also gets a forecast based on the slope of your recent runs. The point of Goals isn't reporting; it's forcing a conversation with content and PR about which two or three numbers you actually intend to move this quarter.

Narratives: how is one specific story landing?

Narratives is the topic-scoped variant of perception. Instead of asking "how is AI describing the whole brand," you ask "how is AI describing this specific topic": a launch, a controversy, a competitive claim, a corrective message you published last month.

Each narrative gets its own score, its own provider breakdown, and its own arc over time. Add a corrective message with the key phrases you want AI to adopt, and Narratives tracks penetration: the percentage of models now repeating your corrective phrasing. It's how you prove a comms campaign moved the needle in AI, not just in earned media.

Use Narratives for: crisis recovery, product positioning launches, fact-corrections, or any time you've published a deliberate counter-narrative and want to know whether AI is repeating it back.

Acting on it

The hardest part of perception data is not reading it; it's deciding which signals warrant a response. A few patterns worth knowing.

Sentiment or category score drops

If a category score drops more than five points in a week, that's something a content campaign didn't cause. Go to the Story tab, look at common concerns and Semantic DNA. A drop almost always shows up as a new descriptor (or a new concern) that wasn't there last week. From there:

• If the source is a single critical article or thread, the source influence footer will name it. Decide whether to respond directly or counterweight with new content.
• If the descriptor is broad and diffuse ("expensive," "complex"), it's a positioning problem you have to address in your messaging, not a single piece to push back on.
• If the drop is in one model only, it may be a training-data refresh on that model's side. Check whether other models confirm it before reacting.

Themes are off

The Semantic DNA reads like a different brand from the one you want. Maybe you're "scrappy" and "startup-friendly" but you've spent a year repositioning for enterprise. The fix is content-led: AI is averaging the public material it can find about you, and right now the average is wrong. Publish enterprise case studies, change your own site copy, get coverage in publications AI cites for the new positioning. Then watch the emerging descriptors list for the new words to start showing up.

A competitor pulls ahead on a dimension you care about

The Competitors tab will name the gap. Open Story for both brands and compare descriptors; the word the competitor owns is usually the word you need to either reclaim or differentiate against. This is the dimension to wire into Goals.

Plan access

Client portal users see the Overview tab only; Goals and Narratives are agency-managed and hidden from end clients.

Common questions

Why is my perception score different from my visibility score?

Different question entirely. Visibility measures whether AI mentions you on category prompts. Perception measures how AI describes you when asked directly about your brand. You can have high visibility and low perception (mentioned often but described poorly), or low visibility and high perception (rarely surfaced but glowing when you are). Both matter, for different reasons.

Why do the four models disagree so much?

Because they're trained on different snapshots of the web, with different reasoning styles, and Perplexity reads the live web on top of that. A 10-15 point spread between models on a single dimension is normal. Larger spreads are themselves signal: a model that diverges hard usually has a specific source driving it, and the source influence footer will often surface what.

How often does perception refresh?

Once a week. Each run replaces the prior week's snapshot for the dashboard view, but every historical run is preserved in the trend chart and the Story history, so you can watch drift over months.

Does adding or removing competitors change my scores?

Yes, materially. Models are told to score brands relative to the others in the batch. Drop a strong competitor and your relative position can tick up; add a category leader and your scores can compress. Don't gerrymander the list; track the brands you actually compete with.

Why is my "Value for Money" score low when we're not that expensive?

Because AI is averaging public discussion, not your pricing page. Reddit threads complaining about price, review-site bullet points that lead with "pricey," competitor copy framing you as premium; all of it gets folded in. The Story tab will usually name the specific concern in plain text. Fix the perception, not just the price.

Can I add my own custom dimensions?

Not in the standard twenty. The categories and dimensions are fixed across all brands so that scores and percentiles are comparable. For brand-specific questions, use a Narrative; that's the custom-topic lane.

Why does Perception need its own setup wizard?

When you first enable it, the wizard asks you to confirm the competitors you want to be compared against and what perception attributes matter most to you. The competitor list drives the batch prompt. The attribute selection drives Goals defaults and the perception-gap analysis on the Story tab. Both can be edited later from the Configuration panel.