The AI shelf
Buyers ask an assistant what to buy. Find out what it says.
Delphi asks ChatGPT, Gemini, Claude, Perplexity and Grok the questions your buyers actually ask, three times each, and reads every answer. You get one number, the five gates behind it, and the specific work that moves the lowest one.
Measure your brandSee an example dashboard
No card to start. The example workspace is clearly labelled as illustrative.
Five gates, one Index
A buyer passes through five gates before they choose you. The Delphi Index weights them, and the lowest gate is your binding constraint: the one that caps everything downstream of it, and the one the plan leads with.
Gate 1 · weight 0.15
FoundCan AI systems access and read the brand at all
Gate 2 · weight 0.25
SeenDoes the brand appear in AI answers for category questions
Gate 3 · weight 0.2
TrustedIs it described accurately and positively
Gate 4 · weight 0.25
ChosenIs it recommended at or near the top
Gate 5 · weight 0.15
BoughtCan the purchase actually complete
How it works
- 1Delphi drafts your question setThe questions a buyer in your category and market would actually type. You edit them before anything is measured, because a set you never saw would make every number afterwards unaccountable.
- 2Five assistants are asked, three times eachThree samples because these models are not deterministic. The spread across them becomes the confidence band that travels with your Index.
- 3Every answer is stored verbatimNot a summary, not a score with the working thrown away. You can open any result and read the exact text the assistant produced, with the model that produced it named beside it.
- 4The plan follows the lowest gateEach item names the mechanism it works through and the range it could move. Mark one shipped and the next run measures whether it worked, and says so either way.
What Delphi will not do
- Invent a number. Where a figure is unavailable, Delphi says so. A question with no licensed demand figure reads as pending, never as zero.
- Hide a partial measurement. If an engine fails, the run completes and states how many of the five actually answered. It never backfills from a previous run.
- Score you on a claim it cannot show you. Every gate traces to answers you can read in full.
- Promise a revenue figure. The plan gives a range in Index points and the basis for it. Delphi measures what assistants say, which is upstream of a sale and not the same thing.
Plans
A credit is one full measurement of one brand: your question set across five assistants, three samples each, plus every competitor scored on the identical set.
One brand, measured properly.
$99a month
- 2 workspace credits a month
- 5 competitors per workspace
- 3 seats
A portfolio, with the plan attached.
$399a month
- 9 workspace credits a month
- 5 competitors per workspace
- 10 seats
- Weekly light re-measure
Clients, each with their own set.
$1,499a month
- 18 workspace credits a month
- 10 competitors per workspace
- Unlimited seats
- White-label reports and read API
Enterprise adds SAML, the reasoning pass and a larger credit allowance. Checkout is being wired up: for now, start an account and we will set the plan by hand.