How to check whether AI answers mention your brand

Ask the questions your buyers ask, in the AI products they use, and record three things per answer: whether your brand is named, whether your domain is cited as a source, and which competitors appear instead. Run the same prompt set on a schedule so the numbers are comparable over time. A single spot check tells you almost nothing, because answers vary between runs.

Why is checking AI visibility harder than checking rankings?

A Google ranking is a stable, ordered list you can look up. An AI answer is generated per request, so two people asking the same question in the same product can get different brands named. That makes ad hoc checking misleading in both directions: one flattering answer proves nothing, and one bad answer is not a trend.

Three consequences follow:

• Use a fixed prompt set, not whatever question comes to mind that day. Changing the prompts changes the numbers.
• Measure rates across many prompts and runs, not the outcome of a single answer.
• Keep the history, because the useful signal is the direction of the rate and which competitors move with it.

What should you measure?

Four signals cover most of what matters, and each one is separately actionable:

• Mention rate: the share of prompts where your brand is named at all. This is the headline number.
• Citation rate: the share where your domain appears as a linked source. Mention without citation means the model knows of you but is not sending traffic.
• Position and sentiment: where in the answer you appear (first recommendation or a footnote) and how you are characterized.
• Competitor share: who gets named when you do not. This turns a bad score into a target list.

Machine readability sits underneath all four: if AI search crawlers are blocked in robots.txt, citation rate has a ceiling no amount of content fixes will lift. Check that first, because it is the cheapest thing to be wrong about.

Which prompts should you use?

Use the questions a buyer would actually type, not your brand name. Brand-name prompts flatter you: an engine asked about "HuskyScout" will discuss HuskyScout. The prompts that reveal visibility are the ones where you have to earn the mention:

• Discovery: "best tools for measuring AI search visibility"
• Recommendation: "what should a solo founder use to audit SEO"
• Comparison: "X vs Y for content teams"
• Alternatives: "alternatives to X"
• Brand: "what is X" (useful as a control, and to catch wrong descriptions of your product)

Keep the set stable. Generating fresh prompts every run makes the trend line meaningless, which is the most common self-inflicted measurement wound.

How do you automate this?

Manual checking breaks down fast: every prompt has to be run in every engine on every schedule, and results have to be recorded the same way each time. Automation is a matrix of prompts times engines, run on a schedule, with the parsing done consistently.

The pieces are: an API per engine (with web search enabled, since the browsing path is what you are measuring), deterministic parsing of each answer for brand mentions and cited domains, and storage so successive runs are comparable. Our AI searchability probe runs that matrix across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, scores it with a published rubric, and keeps the history. If you would rather build it yourself, the design above is the whole shape of it.

What do you do with a bad score?

Fix the retrievability floor first, then the extractability of the content:

• Unblock AI search and live-fetch crawlers in robots.txt, then confirm with a crawler check.
• Publish answer-first content on the questions where you are invisible: the direct answer in the first paragraph, question-style headings, and a real FAQ block.
• Add structured data (Organization, Article, FAQPage) so machines get unambiguous facts instead of inferring them.
• Re-measure with the same prompt set. Movement usually shows up over weeks, not days, because the underlying indexes refresh on their own cadence.

Frequently asked questions

How often should I re-run the prompts?

Weekly is enough for most sites, and it matches how fast AI search indexes tend to refresh. Daily runs mostly measure answer variance, and monthly runs miss the moment a competitor overtakes you.

Does asking ChatGPT about my brand improve my visibility?

No. Prompting a model does not train it, and your conversations do not feed back into what other users are told. Visibility changes come from what the engines can crawl and retrieve.

Why does my brand appear in one engine but not another?

Each engine has its own index, its own crawler set and its own retrieval behavior. Being blocked or absent in one is a per engine problem, which is exactly why the measurement is per engine.

Is AI visibility replacing search rankings?

Not replacing, adding. Classic rankings still drive most discovery for most products, and the same content and authority work feeds both. AI visibility is a second surface with its own failure modes, and the failure modes are the part rankings will not warn you about.