# AI Category Answer Test Protocol

Canonical page: https://waitlist.crany.ai/marketing-templates/ai-category-answer-test-protocol

By Tangenix. Published 2026-10-24. Category: AI search.

Test how search-enabled AI answers a specific product-category decision, checking suitability, factual conditions and cited evidence under a fixed protocol.

## Task

Evaluate the quality and suitability of category answers rather than only counting mentions of a brand.

## When to use

Use this when the team needs to understand whether AI answers help the intended reader choose an appropriate solution. Predefine the task and factual criteria before observing engines or brands.

## Choose another approach when

You plan to keep prompting until your brand wins, infer universal visibility from one answer, or count an inaccessible or gated session as a negative answer.

## Inputs

- A realistic category question and the reader’s actual task or constraints.
- Verified public product and alternative facts needed to assess the answer.
- A fixed observation protocol with engine, mode, locale and account context recorded.

## Template

    Category decision question: [exact stable wording]
    Reader task and necessary conditions: [what a suitable answer must address]
    Expected factual checks: [availability, scope and important exclusions]
    Engine / visible model or mode / observation time: [record actual context]
    Locale / signed-in state / prior-conversation context: [no private account identifiers]
    Search and source access: [available, disabled or blocked]
    Answer observation: obtained / gated / failed / not tested
    Suitability assessment: [which task constraints the answer handled or missed]
    Material claim checks: [supported / contradicted / unresolved with public evidence]
    Citations: [actual links and whether they support the associated claims]
    Alternatives and conditions: [faithful summary, not a brand-win score]
    Repeat plan: [same protocol; keep each observation]
    Interpretation limit: [sample, account context and system variation]
    A missing answer is missing evidence; an on-topic answer without your brand is not automatically poor quality.

## Illustrative example

This is not a Crany customer result. Replace example details with verified facts.

    Fictional example: a founder asks a search-enabled assistant about software for coordinating a small client handoff. The protocol requires checking current availability and whether the suggested tools address the actual task.
    An answer describes a waitlist-only product as generally available. The reviewer records that specific factual error against the public access page rather than celebrating the brand mention. Another answer suggests a suitable manual approach and explains its limits; it may be useful even if the founder’s product is absent.
    If a service requests account verification before returning an answer, the row is marked gated. The reviewer does not invent a citation result, bypass the gate or change the question repeatedly to obtain a favorable screenshot.

## How to use

1. Define the category task and factual suitability checks before running the question.
2. Record actual engine context, answers and supporting links without selecting only favorable attempts.
3. Evaluate correctness and fit, preserving gates, uncertainty and each repeat observation.

## Checks

- A mention or citation is not assumed to be a correct recommendation.
- An authentication gate is not recorded as proof of brand absence.
- The prompt is not altered after seeing results while calling the observation comparable.

## What to measure

Report observed answer suitability, factual errors and claim-supporting citations within the recorded sample. Keep failed or gated attempts visible and distinguish them from obtained answers. This protocol does not estimate global citation share or prove that an SEO edit caused a recommendation.

## Research and sources

Google describes variation in AI answers and supporting links and the role of ordinary search eligibility. This original category-quality protocol is not an official platform benchmark and does not generalize one engine’s behavior to all AI products.

- [Google: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) — checked 2026-10-24. Search eligibility, visible content, AI feature reporting and the absence of special AI markup requirements.

Prepared with AI assistance. Sources support the stated principles; they do not endorse Crany or guarantee performance.

## About Crany

Crany is an AI marketing co-founder in development for software founders and vibe coders. It is designed to strategize, execute approved marketing and improve based on results. These free resources are usable independently of app access. The public waitlist is at https://waitlist.crany.ai/.