Methodology

Platforms we track

SearchIntel monitors brand visibility across six AI search surfaces:

• ChatGPT — OpenAI conversational search
• Claude — Anthropic conversational search
• Gemini — Google's standalone AI chat product
• Perplexity — AI search engine with citation-first output
• Google AI Overviews — the AI summary shown above classic Google results
• Google AI Mode — Google's dedicated conversational search tab, reached 1B monthly active users at I/O May 2026

Each platform draws from different sources, applies different ranking, and produces different answers for the same query. We measure all six because optimising for one does not transfer to the others. Industry estimates put citation overlap between Google AI Overviews and Google AI Mode at roughly 14 percent, so they have to be tracked separately.

How we run a visibility scan

For every query in a client's prompt set, we send the query to each of the six platforms through their official APIs or vendor-backed scraping (SerpAPI for AI Overviews and AI Mode). For the four conversational platforms (ChatGPT, Claude, Gemini, Perplexity) we run multiple repetitions per query because LLM outputs are probabilistic. AI Overviews and AI Mode are queried once per run because their output is deterministic for a given query and region.

For each platform response we capture:

• The verbatim AI answer text
• Whether the client's brand appears in the answer
• Every other brand named in the answer (the competitive consideration set)
• Every URL or source cited (the citation pool)
• The region context the answer was generated under

Source hierarchy

When we cite external evidence in reports or content, we apply a source hierarchy:

1. Primary data we collected ourselves from the SearchIntel scan engine, with the date and scan ID recorded.
2. Official platform documentation (OpenAI announcements, Google blog posts, Anthropic system cards) when describing platform behaviour.
3. Named industry research with the underlying study or dataset linked (Profound, Peec, BrightEdge, SimilarWeb, SEO Clarity, when the methodology is public).
4. Trade press only as a secondary signal pointing at a primary source we then verify.

We avoid citing aggregator pages that compile stats without linking to the underlying studies. If a statistic appears repeatedly across the web but the original source cannot be found, we mark it as unverified and look for a primary-source equivalent.

Confidence labels

Client-facing findings carry a confidence label:

• High confidence — a finding backed by our scan data plus at least one independent corroborating source, observed across multiple runs.
• Directional — a finding consistent with our scan data but based on a single run or a small sample; useful for indicating a likely pattern that needs more data to confirm.
• Observation — a single data point reported as-is, with no claim that it generalises.

We do not present directional findings or observations with the language of certainty. If we are still gathering data on a question, we say so.

Corrections

When we publish something that turns out to be wrong, we correct it visibly:

• The corrected version replaces the original at the same URL.
• The post receives an updated dateModified stamp.
• A short changelog note is added under the title describing what changed and why.
• For substantive corrections (factual error, retracted claim, withdrawn data), a correction note remains visible on the page indefinitely.

If you find a mistake in anything we publish, please tell us at paul@searchintel.tech. Corrections we accept are shipped within five working days.

What we don't claim

SearchIntel is a measurement and advisory agency, not a platform. We do not control how AI systems generate their answers. Our role is to measure what they say, identify the patterns, and recommend changes that have a reasonable chance of moving the result over time.

We do not guarantee specific score improvements within specific timeframes. AI search visibility depends on factors outside our control: model updates, training data changes, third-party citations we don't manage, and the competitive content moves of other brands in the same category.