Human-verified evidence, or nothing
AnswerLift is an evidence ledger for AI answer visibility. A Sprint asks your client’s buyer questions of the models behind ChatGPT, Claude, Gemini and Perplexity — through their official APIs, with the OpenAI, Anthropic and Perplexity models searching the web and Gemini answering from model knowledge — plus five open-API models for comparison. Each answer is stored verbatim with the exact model, the sources its search cited and the time. Your team adds what it saw by hand in the consumer apps. Then a named reviewer verifies every record — collected or hand-entered — before it can appear in a client report.
No estimated scores, anywhere. Collection runs only through official paid APIs, never by scraping a consumer interface, and every record says which engine, model and search method produced it. Where collection is off, the product says so instead of substituting a guess.
Worked example · what a Sprint report shows
One prompt, four major engines, one verified record
Fernhill Accounting (fictional) · fernhill.example
“Which accounting firms would you recommend for a UK e-commerce seller?”
Named by 2 of 4major engines
Counted from stored answers. No score.
| Engine | Named? | Position | Sources |
|---|---|---|---|
| ChatGPTOpenAI GPT API (web search) | Yes | #3 | Own site cited · 1 of 5 sources |
| ClaudeClaude API (web search) | No | — | Not cited · 0 of 4 sources |
| GeminiGemini API (assistant panel) | No | — | No search · model knowledge |
| PerplexityPerplexity Sonar API | Yes | #2 | Own site cited · 1 of 6 sources |
The answer named three firms (Larkspur & Co, Brightwater Partners and Oakden Advisory — all fictional) and cited four directory and review pages; none on fernhill.example. The stored answer is kept in full for the reviewer and the client to read.
Verified by a named reviewer before it could appear in the client report
Records awaiting review are excluded from the client report by the data query itself. Each engine row is the model behind that assistant, asked through its official API — not a screenshot of the consumer app — and every record says which model and which search method produced it.
The differentiator
Every observation — whether a person typed it or a provider API returned it — starts life pending. A person reviews it against the stored answer text, puts their name to it, and marks it verified or needs-review. Only verified records are returned to the client report, and the exclusion is enforced in the data query, so an unverified record cannot reach a client screen even if someone rewrote the page.
Ask the same question again next month and it is stored as v2 of that prompt, alongside v1. Nothing overwrites, nothing is inferred, and a QA decision never rewrites the snapshot it was made about.
Recorded
Prompt, engine or provider model, cited sources, capture time, observation time
Classified
Cited · mentioned but not cited · not mentioned, plus confidence and volatility
Signed off
Named reviewer, review timestamp, and only then client-visible
What the software will not do
Choose the next right step
Available now
One brand, three buyer questions that do not name it, put to the models behind ChatGPT, Claude, Gemini and Perplexity. A person checks every answer, and you get back what each engine said and which sources it cited. Free, with no obligation.
Request SnapshotAvailable now · £2,750 one-time
Start a focused engagement to structure the evidence, priorities, handoff, and practical next actions.
See Sprint scopeFree · no subscription
Record and review evidence by hand at no cost. Automated answer-engine collection runs as part of a Sprint.
Create a free workspaceBuilt for agency handoffs
Evidence is more useful when it carries scope, uncertainty, ownership, and review status with it—not when it is trapped in another dashboard.
Explore the product workflowScopes the prompts, markets, and client questions worth observing.
Receives an assigned remediation brief with the evidence and approval context attached.
Can explain what was observed, what is still uncertain, and what happens next.
Receives a client-safe report containing only human-verified observations.
Straight answers
