How CHOIVE works.

1. What you provide

You enter a business name, website, category, and location. You may also add a differentiator, known competitor, and customer language. CHOIVE uses this information to identify the correct business and create buyer-style questions.

2. What CHOIVE checks

Each new diagnostic checks the submitted website and current public evidence. Depending on availability, this can include page content, structured data, search results, reviews, citations, press, partnerships, client evidence, and technical discoverability.

Earlier provider answers and competitor decisions are not reused as the answer to a new diagnostic. Previous completed runs can be retained for audit and progress comparison.

3. The four AI measurements

CHOIVE sends independently measured recommendation questions through official APIs associated with these consumer-facing products:

CHOIVE adds new AI platforms as they gain buyer adoption, so the measurement always reflects where real buyers are asking questions.

Each provider records one response for each unbranded visibility question. The branded replacement question normally records three independent responses. CHOIVE shows a company only when at least two responses completed and the same name received more than half of those completed responses. Failed responses remain failures. The four provider results remain separate and are not merged into one vote.

An API response is a real provider response for that diagnostic run. It is not a guarantee of the answer every person will see inside a consumer app. Consumer apps can use different models, search context, product settings, conversation history, location, or personalization.

4. The CHOIVE Index

The CHOIVE Index is a score from 0 to 100. It is the sum of four pillars worth 25 points each:

The pillar scores use evidence about the subject business. Whether an AI provider recommended or omitted the business does not automatically raise or lower any pillar score.

Final points are allocated by a versioned evidence rubric. Code converts recorded evidence statuses into points; a model cannot submit its own final score. Mechanical checks are labelled mechanical, independently retrieved sources are labelled independent, and interpreted checks are labelled model assessed. Clarity and Difference include interpreted checks and therefore carry lower confidence when independent verification is limited.

Every paid Analysis and Report includes a point ledger showing the rule ID, observed evidence, points awarded, maximum points, verification type, source where available, and pillar confidence.

5. How CHOIVE protects the measurement

A recommendation diagnostic is only as trustworthy as the discipline behind it. CHOIVE applies the same safeguards to every run, including its own, so a score reflects evidence, not noise, guesswork, or manipulation.

Evidence-first hard limits, not model opinion

Some points are decided by mechanical rules the language model cannot override. For example, if the Ease pillar is being scored and the site has no structured data, Ease is capped low; if verified structured data is present, Ease has a guaranteed floor; a published llms.txt file raises that floor further; and if the page is an empty shell that AI crawlers cannot actually read, Ease is capped near the bottom regardless of any metadata claim. Code enforces these limits after the model responds, so no assessment can quietly inflate a weak signal.

Consensus before a name is shown

The branded replacement question records several independent responses. A company is only shown when at least two responses completed and one name received a strict majority of them. A single lucky mention is not treated as a recommendation, and failed calls stay failures rather than being filled in.

Ground truth versus projection

Signals that are mechanically verified are labelled as confirmed ground truth. Where CHOIVE illustrates how stronger evidence could change an answer, that paragraph is clearly marked as a projection, never presented as a recorded provider response.

Protection against manipulated inputs

Text pulled from a website or search results is sanitized before it reaches the model, so instructions hidden inside a page ("ignore previous instructions", injected prompts) cannot steer the diagnostic. Off-topic community chatter (entertainment, gaming, and hobby forums unrelated to the business category) is stripped out and can never be counted as a trust signal.

No manufactured superlatives

Every ready-to-use asset and written finding is checked against a banned-claims filter. Words like "best", "leading", "market-leading", "trusted", "award-winning", or "number one" are not allowed unless a credible independent source in the collected evidence proves that exact claim. CHOIVE describes an evidenced difference precisely instead of inventing one.

Buyer questions matched to the business model

The recommendation questions are generated to match how a real buyer of that specific type of business would ask (the phrasing for a local service differs from a SaaS product or a marketplace), so the measured answer reflects a realistic selection moment rather than a generic query.

6. Competitor roles

AI recommendation

The company a named AI provider recommended during the recorded query. Each provider can return a different company.

Head-to-head competitor

The closest verified purchasing substitute: a business selling a comparable product or service to the same buyer type, in a market the subject can actually serve.

Wider market competitor

A relevant competitor in the broader category that may not be the closest like-for-like substitute.

CHOIVE does not describe a market competitor as an AI recommendation unless the recorded provider response names it.

7. What the three products contain

• Free Diagnostic, $0: headline verdict, CHOIVE Index, four pillar scores, platform recommendations when successfully measured, and competitor exposure.
• CHOIVE Analysis, $99 once: the full online evidence breakdown, platform coverage, competitor comparison, prioritized actions, and ready-to-use assets.
• CHOIVE Report, $499 once: the Analysis plus a structured ten-section report, implementation milestones, verification criteria, and a 30-day plan. It does not predict future scores.

8. Failures, uncertainty, and missing evidence

If a provider request fails or an answer cannot support a reliable company name, CHOIVE should show the measurement as unavailable, partial, or not established. It should not silently invent a recommendation.

A low-confidence competitor must not be presented as a confirmed recommendation leader. A missing public trust signal means CHOIVE did not verify that signal during the run; it does not prove the signal can never exist.

9. What CHOIVE does not promise

• It does not guarantee that an AI platform will recommend a business in the future.
• It does not guarantee an API response will match every consumer-app conversation.
• It does not treat the CHOIVE score as a vote from an AI provider.
• It does not guarantee that implementing an action will produce a specific future score or commercial result.
• It does not replace legal, financial, or professional advice.