01
01Customer question intelligenceMeasure the questions that can actually move a sale.Research, comparison and buying questions, structured across every assistant.
Answers
Live AI recommendation intelligence
AI is already making choices for your customers. Make more of those choices yours.
Delphi puts buying questions for your category to every AI assistant, keeps the answers, and tells you exactly what to change.
This is the competitive edge in AI you are looking for.
From $99 a month. Monthly, cancel any time, no trial theatre.
ChatGPT answers
For flat feet you want structured support and a firm midsole. Most people do well with Kestrel Run Arc 4, which is the one I would start with. Ardent Stride also makes a stability shoe, though it gets less consistent feedback for overpronation.
Named, not recommendedSecond, behind a rival it recommends by name.
Five name you and put a rival first. One advises against you. Ardent, Kestrel, Trailhead and Nimbus are invented.
Four ways an answer costs you a sale.
- Never named
You are not in the answer at all.
- Named, not recommended
You are in it. Another brand is recommended first.
- Advised against
You are in it, and the answer steers the buyer away.
- No path to a purchase
Recommended, and the answer names nowhere to make a purchase.
Thirteen layers of intelligence, on one loop.
Each answers a different question about the same run. Together they are the difference between knowing you lost and knowing what to change.
Zone I
The questions that decide the sale, and the ones you are not in.
02
02Loss intelligenceSee the exact questions where demand goes elsewhere.Move from one score into the individual buying moments you win and lose.
03
03Competitive choice intelligenceYour position changes by question and by assistant.Who wins each decision, where assistants disagree, which rivals appear that you never named.
Standing
Zone II
What AI believes about you, the sources it cites, and whether the path to buy holds.
The question changes here: from what is happening, to why it is happening.
04
04Brand belief and claim intelligenceSee what AI believes about you, down to the claim.The attributes, narratives and exclusions attached to your name, with the evidence beneath.
Claims
05
05Narrative diagnosisFind the belief that is costing the recommendation.Not whether sentiment is positive, but which specific idea made you harder to choose.
06
06Source and influence intelligenceTrace what AI believes back to the sources it cites.The domains and pages the assistants cite on your questions, ranked.
Sources
07
07Source prioritisationNot every source matters. Find the ones that do.The publishers that keep recurring around competitive outcomes, rather than the open web as one problem.
08
08Digital shelf and purchase readinessWinning the answer is useless if the path to buy breaks.Whether an assistant can find, understand, quote and route a buyer to somewhere they can actually purchase.
Readiness
Zone III
Why it happened, what moves first, and who has to change something.
The question changes here: from what is true, to what to do.
09
09Root cause diagnosisWe do not stop at you lost. We reconstruct why.The chain from the outcome back to the evidence and the assets behind it.
10
10PrioritisationNot every finding deserves action. This is what moves first.Ranked by what it is worth, how confident we are, the effort, and the part of the journey it moves.
Plan
11
11Prescriptive intelligenceThe outreach is already written. Not a task list, a letter you can send.Per named source: the letter, the reasoning it rests on, the community and the exact thread to join, and the expertise worth bringing to it.
Zone IV
The floor of the section. It needs every layer above it to exist.
The question changes here: from what you did, to what it moved.
12
12External interventionChange the evidence, not just your website.Upcoming: source gaps will become evidence-backed third-party work, with editorial judgement left with you.
LabsOff your own site
13
13Closed loop measurementMeasure what moved, and what was only noise.Runs compared across recommendation, questions, rivals, assistants, claims and sources, with the margin of error drawn around the Delphi Index.
Trends
Then the next run cuts the same section again, against the same questions and the same assistants, and the dashboard and the trends page say whether the Delphi Index moved by more than its margin of error. That is the loop, and it is the only part of this a screenshot cannot fake.
The whole reading, nothing behind a click
One run, end to end.
This is what a run gives you. One illustrative brand, measured the way yours would be.
Ardent · the United States · run 3Hypothetical brand
Delphi Index45Up 2 since run 2
Recommended43%186 of 433 answers
Buying questions lost7of 20 at the point of choosing
Named, not recommended140answers where another brand came first
Rivals recommended144 outside the tracked set
Path to buy40%of answers name a destination
Root causeThe five steps, in orderThe weakest caps the rest
Found71Good
Seen64Good
Trusted7Weakand it caps the four above it
Chosen43Fair
Bought40Fair
Trusted is the lowest step, so the plan starts there: a buyer who does not trust the brand never reaches the steps after it.
Closed loopWhat actually movedSince the last run
Where it wasWhere it is nowThe margin of error
Delphi Index45Within ±8
Found71Up 3
Seen64No change
Trusted7No change
Chosen43Up 4
On the dashboard and the trends page, a change in the Delphi Index smaller than its margin of error is not called a movement.
Competitive choiceEvery question, every assistantSame market, different answers
RecommendedNamed, not recommendedAdvised againstNever namedNo AI answer
| Buying question | ChatGPT | Google Overview | Gemini | Copilot | Claude | Perplexity | Grok |
|---|---|---|---|---|---|---|---|
| Best running shoes for flat feet? | ChatGPT: Named, not recommended | Google Overview: Named, not recommended | Gemini: Never named | Copilot: Never named | Claude: Named, not recommended | Perplexity: Advised against | Grok: Named, not recommended |
| Ardent Stride or Kestrel Run Arc for overpronation? | ChatGPT: Never named | Google Overview: No AI answer | Gemini: Named, not recommended | Copilot: Named, not recommended | Claude: Never named | Perplexity: Recommended | Grok: Never named |
| Which stability shoe has the widest toe box? | ChatGPT: Recommended | Google Overview: Recommended | Gemini: Recommended | Copilot: Named, not recommended | Claude: Named, not recommended | Perplexity: Recommended | Grok: Named, not recommended |
| Cheapest stability shoe that actually lasts? | ChatGPT: Named, not recommended | Google Overview: Never named | Gemini: Never named | Copilot: Never named | Claude: Named, not recommended | Perplexity: Never named | Grok: Advised against |
SourcesWhat they cited1,364 citations
reddit.com72
runnersworld.com47
wirecutter.com46
ardent.comYour site45
believeintherun.com43
Brand beliefThe words attached to you, and to your rivalsTracked rivals own 7 words you do not
Only you10
roomy toe box12
hard-wearing outsole4
true to size3
best for long runs3
Only them7
reliable stabilityKestrel9
what most people are toldKestrel9
averageTrailhead6
fine if you also walk in themNimbus5
Both of you1
best overall5
One word in common is the whole overlap. Seven belong to somebody else.
A claim is listed only where the answer text is stored beside it. We quote both sides and do not rule.
PrescriptiveThe work, in order22 items, 3 land quickly
01Correct the arch support line on wirecutter.comTrusted, 3 answers cite this page against youEvidence backed
02Publish a flat feet fit guideTrusted, answers the question you lose mostEvidence backed
03Name a stockist on every product pageBought, 60% of answers name no pathInferred
04Take part in the thread the assistants cite mostSeen, the community is the most cited sourceHypothesis
ScopeWhat every run disclosesOn the run itself
1How many answers it was scored onRun 3 was scored on 433 of 486 planned answers.
2Which assistants answered only partiallyOne returned 19 of 72. The gap is printed, not averaged.
3When a change is not a movementOn the dashboard, a change in the Delphi Index inside its margin of error is left uncalled.
4When it could not tellThree steps report no margin this run, and say so.
5Where Google showed no AI answerGoogle showed no AI Overview on 6 of 24 questions. They are left out of the figures for Google Overview, not counted as unnamed.
A disagreement about your brand is a finding, not an error to average away.
Ardent, Kestrel, Trailhead and Nimbus are hypothetical brands, used to show the shape of a run. Delphi does not put a real customer's measurement on a marketing page.
Find out what the assistants tell your buyers.
Buying questions for your category, put to ChatGPT, Google Overview, Gemini, Copilot, Claude, Perplexity and Grok, with the plan to fix what they get wrong.
- Starter$99
- Scoring and the plan. One brand, email support.
- Growth$399
- Scoring and the plan. 3 brands, office hours support.
Agency and Enterprise carry more brands and a second run a month for each brand. Every level is monthly and cancellable from your own settings.