Authority is a position, not an opinion. Position determines who buyers see first.
Your buyers are asking AI which company to choose, and AI is answering today. The ARDI™ Brand Index shows you exactly where your brand sits in those answers, which competitor is being named instead, and the specific moves that change it. Below is a condensed walk through a real report. One half of the ARDI™ Authority Index inside GOSH AI, the AI Recommendation and Discovery Intelligence platform.
What matters now.What is the ARDI™ score?One composite out of 100, and it is the headline of the whole report. It blends Discovery Visibility, how often models surface you unprompted, with Recognition Visibility, how well they know you when asked. The three smaller figures beside it are the diagnostics that explain the score. They are never averaged into a second number.
One number a board can act on, and the two measurements that explain how you got it.
Full Authority. ARDI™ Score 39.
Aurora Metrics is in the Emerging Score Tier and the Full Authority Authority State. The Authority State gap is closed.[3]
Trusted, and measured.
Four states are possible. The one you are in decides whether your problem is proof, visibility or existence, and stops you spending a year fixing the wrong one.
Eight surfaces, and they do not agree.
The gap between the two dots is the distance between being known and being chosen. Win one surface and you have won one eighth of the market.
Google AI Overviews is measured and reported like every other surface. It appears on a minority of questions, so it is the one surface kept out of the composite rather than averaged in. Point-in-time results vary; outlier is a measured classification using an explicit 15 point distance rule, not a judgment.
The journey breaks at the moment of choice.How do I read the heat map?One row per buying question, one column per AI surface, and a cell for what that surface did with you on that question. The 50 questions are split across the three buyer stages: Think, Discovery and Decision. Reading down a column shows one surface’s behaviour. Reading across a row shows how much the surfaces disagree on the same question.
Deals are won and lost at different stages. This shows the exact stage where yours are being lost.
| Stage | Prompt | ChatGPT | Copilot | Perplexity | Claude | Gemini | Grok | AI Mode | AI Overviews |
|---|---|---|---|---|---|---|---|---|---|
| Think | Think prompt 01 | REC | MEN | · | · | · | · | REC | · |
| Discovery | Discovery prompt 01 | · | · | REC | MEN | · | · | MEN | · |
| Discovery | Discovery prompt 13 | · | · | · | · | · | ABS | · | ABS |
| Decision | Decision prompt 01 | · | · | · | · | ABS | · | ABS | · |
| Decision | Decision prompt 10 | · | · | · | · | · | AMB | · | · |
Representative rows from the complete 50 prompt audit. The full map runs every prompt against every measured surface.
Someone is winning these answers. It is not you yet.
The competitors taking the answers you are absent from, named, with the questions worth taking back first.
When buyers asked which one to pick, every recommendation went to a competitor. None went to Aurora Metrics.[2,23]
Observed evidence establishes occupancy of Decision-stage consideration. It does not establish causality or future outcome.
The sources deciding whether you get named.Why do citations matter?Models weight sources they already trust above anything a brand says about itself. Every cited domain is checked against a verified ownership registry, so you can see which support is independent, which is your own, and which belongs to a competitor. A source nobody has verified is a finding in its own right.
Most of them are not yours. This is who AI already trusts in your category, and which sources are worth going after first.
One cited domain does not resolve in the verified ownership registry. Models are reading a source nobody has verified.
Where the 18 citations sit in the journey: Think 3, Discovery 12, Decision 3. By surface: ChatGPT 5, Claude 4, Perplexity 4, Gemini 3, Google AI Mode 3, Google AI Overviews 2, Grok 1.
The support exists. The chain leaks.What is public footprint conversion?Your public footprint is everything that should carry your brand beyond your own site: profiles, listings, directory entries, third party pages, and the links between them. Each one is tested against five questions in order. Does it exist, is it complete, do its facts agree with your other sources, is it connected back to you, and did AI actually use it. An asset that fails one question stops there and counts for nothing, however good it looks to a person. Conversion is how many make it all the way through.
Every asset that stalls is a place a competitor gets named instead. Twelve public assets, five questions asked in order, and the exact point where the chain breaks.
What the models get wrong.What counts as a misattribution?A case where a model states something about you that conflicts with verified evidence. Each one records what was observed, what the correction is, and a severity. The audit proves the conflict exists and deliberately stops short of claiming it caused the model’s behaviour, because that is an assertion the evidence does not support.
What AI says wrongly about you spreads without correction. These are the errors worth fixing before they harden.
Three moves carry most of the plan.
Ranked by evidence and by what you actually control, so the first thing you do is the thing that moves the most.
| # | Action | Priority | Why it is first |
|---|---|---|---|
| 1 | Create Build the page that compares you to the alternatives | Score 81 | A verified absent comparison asset sits alongside a measured Decision-stage gap. |
| 2 | Correct Fix the incorrectly listed product category | Score 78 | The public category conflicts with approved identity evidence. |
| 3 | Unify Make every source describe the product the same way | Score 74 | Disagreeing sources give models two versions of the same brand. |
Who does what, and who must say yes.
Seven actions sit with six teams. Nothing ships without client approval.
| Action | Does it | Supports | Says yes | When |
|---|---|---|---|---|
| Build the comparison page | Content | Website or technical | Client approves | Days 31 to 60 |
| Fix the product category | Shared | Legal or compliance | Client approves | Days 0 to 30 |
| Unify the product description | Website or technical | None | Client approves | Days 0 to 30 |
A sequence, not a sprint board.Why does the order matter?Because the phases depend on each other. Corrections have to land before new content is worth building, and content has to exist before a third party will cite it. Running them in parallel wastes the first two. Every item carries an owner, an approval, a deliverable, and the checkpoint that will prove it worked.
Ninety days is the window where movement shows up in the next measurement, provided the work runs in order.
Where the brand goes next is a decision, not a promise.
The target state is a commitment with conditions attached. You approve it, or it does not exist.
How you will know what actually changed.
The same questions, the same surfaces, the same rules. What moved is attributable, not asserted.
Trust, then verify. Everything is here.
The main report is selective. The appendix is complete.
50 buying questions across Think, Discovery and Decision, written the way buyers write them. Your brand is never mentioned, so any appearance is the model’s own choice.
Every numbered reference resolves to the response that produced it. Nothing is asserted without the evidence behind it being retrievable.
Stated in place, not buried. Where evidence proves a conflict but not its effect, the report says so rather than implying causality.
Measure. Move. Re-measure.
That is how a position changes. The ARDI™ Brand Index measures whether AI knows your brand well enough to name it unprompted. The ARDI™ Product Index measures whether AI then names one of your products, or a competitor’s instead. Two halves of one measurement. The authority half of GOSH AI, the AI Recommendation and Discovery Intelligence platform.