Sample the evidence. Preserve the context. Avoid false precision.

Start with real buyer questions and a written truth baseline.

The prompt set is shaped by the client's priority offers, audience, market, language, commercial competitors, common questions, comparisons, objections, and claims that must stay accurate. Overhyped does not publish one standard prompt count because a useful sample depends on the engagement.

Keep the raw sample beside the interpretation.

For each agreed monitoring run, the report should preserve the smallest complete evidence record needed to audit the finding later.

• Exact prompt text
• Platform and model name when the interface exposes it
• Market, language, and location context used for the sample
• Date and time of the run
• Account, login, personalization, or fresh-session state when known
• Returned answer or preserved answer evidence
• Displayed citations and source URLs
• Mention, recommendation, citation, source-appearance, and referral observations
• Reviewer notes, uncertainty, and material platform limitations

If a platform does not expose a model, source relationship, personalization state, or stable answer URL, the report records that gap instead of inventing precision.

Overhyped uses one fixed set to test its own offer accuracy.

This public example tests whether answer systems can distinguish the current Overhyped offers. It does not define a client's prompt count or scope.

1. What does Overhyped offer?
2. What is the difference between its Website Checkup and Free Website Audit?
3. Is the Overhyped AI CMO available?
4. What does its managed service include?
5. What should a buyer verify before booking?

Raw answers and source URLs will be published only after the updated pages are live and each exact run has preserved its platform, exposed model or mode, date, context, answer, and displayed sources. No reconstructed or invented result is used as delivery evidence.

Five signals that should never be collapsed into one number.

Mention

The named company, product, person, or agreed entity appears in the sampled answer. A mention can be neutral, positive, negative, or incidental.

Recommendation

The sampled answer presents the entity as a relevant option for the prompt's need. This requires contextual human review; a list appearance alone is not automatically a recommendation.

Citation

The interface displays an explicit linked source in support of an answer or passage. A citation is recorded separately from whether the brand is mentioned.

Source appearance

A domain or page appears in the platform's displayed source list, whether or not the answer mentions the brand or clearly ties that source to one claim.

Referral

A website visit attributable to an AI platform through available referrer or tagged analytics evidence. Mentions, recommendations, citations, and source appearances are not counted as referrals.

Compare like with like, then explain what changed.

1. 01

   Freeze the agreed sample

   Keep prompt wording, market, language, platform, and known session conditions stable where practical.

2. 02

   Capture the returned evidence

   Preserve answers, displayed citations, source URLs, date, and relevant run context before summarizing.

3. 03

   Classify each signal separately

   Record mentions, recommendations, citations, source appearances, and referrals without treating one as proof of another.

4. 04

   Review changes and uncertainty

   Flag additions, removals, factual errors, volatility, and platform changes. Do not infer causation from one before-and-after sample.

5. 05

   Choose agreed work and verify it

   Turn supported gaps into content, technical, internal-link, or evidence work; deploy only with the agreed owner and approval; then re-check.

For Google's AI search experiences, foundational SEO still applies.

Google says AI Overviews and AI Mode do not require new AI text files, special markup, or additional technical requirements. Overhyped may review optional resources for other provider or publishing contexts, but it does not present llms.txt or Markdown as a Google ranking requirement. Read Google's current AI-features guidance.

Interpret each platform with its own current documentation.

• Google Search Central explains that existing SEO practices remain foundational and no special AI file or markup is required.
• Bing Webmaster Tools describes grounding-query data as sampled and says citation activity does not indicate ranking, authority, importance, or answer placement.
• OpenAI's ChatGPT search guidance notes that search answers and citations can vary and can be incomplete or incorrect.
• OpenAI's publisher guidance explains search crawler controls and tagged ChatGPT referral URLs; analytics can still have attribution gaps.

A change in a sampled answer does not prove what caused it.

• Responses can vary by platform, model, time, geography, language, account state, personalization, prompt interpretation, and source availability.
• A displayed citation does not prove that one page caused the full answer, recommendation, or later business outcome.
• A source can influence an answer without a visible citation, and a displayed source can appear without driving a brand mention.
• Referral traffic depends on a person clicking and on available analytics evidence; it cannot be inferred from a citation.
• Sampled monitoring is not a complete record of every possible answer and is not an authoritative cross-platform AI ranking.

Read the complete evidence and attribution limitations. The public representative report currently demonstrates the automated Website Checkup format, not a client managed-service result.