Brand Perception · Methodology
How we measure what AI says about you.
Higher score = less drift: the engines describe you the way you describe yourself. Ten weighted dimensions, scored 0–10 against verbatim engine answers.
The 10 drift dimensions
| # | Dimension | Weight | What it scores |
|---|---|---|---|
| 01 | category | 16% | Does AI name your category correctly, or reach for the nearest familiar bucket? 10 = engines name your actual category; 0 = “an AI tool” / “a software platform.” |
| 02 | differentiator_clarity | 12% | Does AI articulate what sets you apart, or stay vague/absent? 10 = engines state your wedge; 0 = no distinctive differentiator surfaces. |
| 03 | target_user | 12% | Does AI name your buyer, or default to “businesses of all sizes”? 10 = the specific ICP your site names; 0 = generic “companies / teams / anyone.” |
| 04 | core_value | 10% | Does AI state the outcome you deliver (e.g. trial→paid), not just features? 10 = the value you sell; 0 = a feature list with no thesis. |
| 05 | credibility | 10% | Does AI cite your own site + credible third parties, or nothing / competitors? 10 = your domain + reputable sources; 0 = aggregators or rivals framing you. |
| 06 | brand_vs_product | 10% | Does AI describe a company, or reduce you to a feature / tool? 10 = a brand with a story; 0 = “a feature inside someone else’s product.” |
| 07 | tone_sentiment | 8% | Is the framing neutral-to-positive, or does a stale knock surface? 10 = fair, current sentiment; 0 = a wrong or outdated weakness that costs deals. |
| 08 | competitors | 8% | Does AI know who you compete with, on your terms? 10 = you lead the comparison; 0 = a footnote to a competitor. |
| 09 | named_customers | 8% | Can any engine name a customer? 10 = engines volunteer real customers; 0 = 0/4 engines can name one (the clearest fix path). |
| 10 | voice_tone_match | 6% | Does AI describe you in your stated voice, or generic-corporate? 10 = on-brand voice; 0 = flattened to boilerplate. |
Composite (0–100) = the weighted mean of the ten values ×10. We ask all four engines six standard questions (rendered with your brand name) and quote a verbatim answer in every score’s rationale. AI engines re-crawl and re-train on a 3–9 month lag, so positioning fixes do not show up overnight.
Cross-engine variance
Variance is its own signal. When the four engines give four different answers to “what does this brand do?”, the brand has no stable machine-readable identity — and the disagreement itself is the story. We report the agreement rate qualitatively (high / mixed / fragmented) with the quotes that prove it.
The zero-brand rate
The cover’s headline number. We run a battery of category prompts — the buyer questions where your category is named but your brand is not — across all four engines. The zero-brand rate is the share of those runs that return no mention of your brand at all:
zero_brand_rate = prompts_with_no_brand_mention / total_category_prompts
We pair it with the same rate for the named category leaders (e.g. “Userflow 8% · Appcues 15% · You 73%”) so the absolute number lands with competitive context. An absolute “73% of category prompts never name you” lands harder than a relative score — that’s why it leads.
Category-king selection
The category-king story strip compares, side by side, how the engines describe youversus how they describe the category king — the one brand the engines name first and describe with a single canonical story. We select the category king as the brand that appears most frequently and most consistently across the comparison-query answers (most first-position mentions, lowest cross-engine variance in its one-line description). We then extract one headline phrase per engine for both you and the king, and footer the strip with each side’s drift (“3 different categories” vs “0 — one canonical story”). The point is visceral: it shows that fixing identity drift is the right first move.
What this audit does not do
- It does not invent engine answers — every score quotes
probe_enginesoutput verbatim. - It does not fill the AEO GEO levers (those are an AEO-only concept).
- It does not promise instant fixes — AI propagation runs on a 3–9 month lag.
- It is a ten-dimension, six-question methodology — not the legacy “5 dimensions / 10 questions” (or the interim 8-dimension) framing.
See also the AEO and CRO methodologies, or browse the audit library.