Signalpoint

How we measure AI visibility

Every report and every study we publish follows this protocol. If a number of ours ever looks wrong, this page is where you check our working.

1. What we are actually measuring

When someone searches a buying question, Google often writes an answer above the results and names specific products inside it. We measure one thing: whether a given brand is named in that answer, and how often, across a set of buying questions for its category.

We are not measuring search rankings, traffic, or sales. A brand can rank first on Google and never appear in the answer above it. That gap is the whole reason this measurement exists.

2. How questions are chosen

Each category gets 25–30 questions, selected against four rules:

• Buying intent, not research intent. The phrasing a person uses when deciding what to purchase.
• Weighted toward recommendation questions. We measured this: "best X for Y" phrasing triggers a product answer naming several brands, while advice phrasing ("can I use X with Y") usually names nobody. Roughly 85% of each set is recommendation phrasing, because that is where visibility exists at all.
• Long-tail over head terms. Head terms are decided by large publisher listicles and are not realistically winnable. Specific questions are.
• A minority of advice questions are kept deliberately — they are where nobody is named, which makes them the clearest openings for a brand to claim.

The full question set is published with every study and included in every client report. We do not keep the questions private.

3. How answers are captured

• Queries are run against Google with India set as the region.
• The AI Overview is captured in full, including the product carousel. Brand names frequently appear only inside carousel entries — reading the prose alone under-reports visibility, and we verified this changes results materially.
• Where Google defers the answer behind a token, we follow it and capture the expanded version rather than recording an empty result.
• Every captured answer is archived with its timestamp and its full source list.
• Queries that return no AI Overview are recorded as such, not silently dropped.

4. How brands are matched

Brand names are matched on the full name with word boundaries, tolerant of spacing and ampersands, so "Dot & Key" and "Dot and Key" both match and "Plum" does not match "plumping". Matching is case-insensitive and applied to the whole captured answer.

We also record every brand-like entity in the answer that is not on the tracked list. This is how competitors nobody was watching get found, and it is included in client reports.

5. What we archive

For every question: the full answer text, every cited source URL, every brand named, the capture timestamp, and whether an AI Overview appeared at all. Clients receive this as CSV. For published studies it is downloadable by anyone, with no email required.

6. What this method cannot tell you

7. Version history

When this method changes, the version number changes and the change is recorded here. Reports state the method version they were produced under.

Questions about the method

If something here is unclear or you think a step is wrong, write to hardik@signalpoint.in. Corrections get published.