A separate human-led service, quoted after scoping: software observations and evidence support a human executive readout. This is not a subscription or a promise of access to every planned intelligence source.
The Diagnostic tells you what to test next, from evidence. A possible engagement examines 20 priority buyer decisions, who is named instead of you, and five intervention hypotheses. We confirm the questions, coverage and deliverables in your written quote.
20 buyer decisions, observed repeatedly, with the raw evidence and who is named instead of you.
Test
Five testable intervention hypotheses, each with the decision it targets and the change. Measurement plan agreed before anything changes.
Verify
Run the strongest hypothesis and re-observe (Choice Experiment Sprint, proposed separately).
The re-observation that verifies it is the optional Choice Experiment Sprint, quoted separately before work.
Who it is for
B2B founders and marketing/growth leaders who need to choose which buyer question to investigate, explain the evidence to stakeholders and scope the next test.
The problem it solves
You need to know where AI answers omit or misrepresent your offer before deciding what to change. The study records what happened, compares alternatives and turns findings into hypotheses to review with your team.
Example engagement scope, confirmed in your quote
20 priority buyer decisions, repeated AI observations, competitor displacement, evidence, five intervention hypotheses, and an executive readout.
20
Priority buyer decisions
Chosen with you by revenue relevance, in customer-approved buyer language.
3
Supported providers, repeated observation
Gemini, OpenAI and Perplexity, 3 runs per question, up to 5 only when a pre-registered trigger fires. Every model version recorded.
Competitor displacement
For each decision, who is named or cited instead of you, and how often. Co-occurrence, not cause.
Raw evidence
Every answer, every cited source. Nothing summarised without its receipt.
5
Testable intervention hypotheses
Each with the decision it targets, the change, and why the evidence points there.
Measurement plan
Windows, sample sizes and comparison decisions, agreed before anything changes.
Executive readout
A live session with your leadership and a report written to be forwarded.
Mention Share per provider: the share of answers that name you. Being named is not being chosen: Choice Share and First Choice are reported only from a classifier that has passed its benchmark gate.
Example engagement sequence
Day 0
Scoping call. You name competitors and share how buyers describe the problem.
Day 1
We draft 20 buyer decisions. You approve or rewrite every one before a run starts.
Days 2 to 4
Repeated observation on every provider, evidence capture, and who is named instead of you.
Executive readout
Findings, five hypotheses and the measurement plan, with your leadership in the room.
Next experiment
Optional Experiment Sprint: run the strongest hypothesis and re-observe. The Experiment Sprint is quoted separately before work.
The written quote confirms the included steps and dates. The dashed step is proposed separately.
Proof
The Diagnostic is new, so there are no client results to show yet, and we will not invent them.
Book the Diagnostic call
15 minutes with Yaniv Goldenberg. We confirm fit, your category and competitors, and the start date. No preparation needed.
Fee
Written scope and fixed fee agreed before work
Readout
Executive session and report
Time zone
Shown in yours
Scheduling runs on Cal.com. Showing it shares your IP address and browser details with Cal.com.
Can I explore ChoiceOrigin before requesting access?
Inspect the public recorded report or request a demo to explore ChoiceOrigin with the founder. A demo request does not create an account, grant product access or start research. The Diagnostic uses your 20 priority decisions in approved language, 3 runs per question (up to 5 only when a pre-registered trigger fires), a competitor analysis and a test plan.
Will you tell us why AI answers leave us out?
We show what appears alongside the answers that do not name you and turn it into hypotheses. No one can see a model's reasons, so we never claim to.
What if a provider is unavailable during the run?
The report marks it unavailable rather than estimating its answers.
Know which decisions leave you out, and what to test first.