Blog / AI & search /
AI chat answers and brand visibility: an early look
Someone sends you a screenshot: your brand, named in a chat answer. No blue links, just a paragraph that mentions your company and maybe cites a page.
Cool. But what does it mean?
In September 2024, generative answers show up in several forms, from Google’s AI Overviews to conversational search prototypes. For most site owners the question isn’t yet how to optimise for a new dashboard. It’s how to understand what you can actually observe.
Brand visibility in AI answers is early and uneven. Systems retrieve and present information differently, and availability varies by product, market and query. A screenshot isn’t the full picture, and appearing in an answer doesn’t mean anyone clicks.
Separate the different products
AI Overviews are part of Google Search and, at this time, had expanded beyond their initial US availability in selected countries. SearchGPT was announced as a prototype by OpenAI in July and was available only to a small group of users and publishers. ChatGPT could also answer questions from its training and tools depending on the product configuration. These are not one shared search engine with a shared ranking system.
That matters when someone says “AI search mentioned us”. I’d ask where the answer appeared, when it was captured, what the query was, whether they were signed in and which sources were cited. Without that, you can’t repeat the observation or draw a business conclusion.
Keep the terminology accurate for the date. Google announced AI Overviews in May 2024; SGE was the earlier Search Generative Experience. Don’t call a test or prototype a universal feature. If access is limited, say so.
What can be measured today
Search Console provides reporting for Google Search performance, not a complete record of every AI-generated response. Google may include some AI Overview activity within existing search reporting, but the available data and its granularity are limited. Analytics can show visits that arrive through known referrals, but some systems may not send a distinct referrer. A click attributed to an AI source may be difficult to identify consistently.
A manual check can answer a narrow question, but it’s no measurement system on its own. Results differ by location, language, personalisation, device and time. If you check a query repeatedly, log the date, market, wording, interface and sources shown. Do not interpret the absence of your brand in one test as evidence that it is invisible everywhere.
I’m wary of third-party tools promising a definitive AI visibility score. Ask how they collect prompts, how often they refresh, which systems they cover and whether you can check the results yourself. A tidy graph can hide a small or biased sample.
Improve the source information first
Brands are easier to represent accurately when their information is clear across the web. Keep the website’s organisation, product and contact information consistent. Make service descriptions specific, use clear page titles, and explain who is responsible for important content. Correct errors in official profiles and reputable third-party listings.
Publish pages that answer real questions with evidence and context. A product page should explain what the product is, how it works and what limitations apply. A service page should define its scope. A technical guide should cite sources and distinguish fact from interpretation. This is useful for readers and gives search systems better source material, though it does not guarantee inclusion in a generated answer.
Don’t make a page for every prompt you can imagine. If it isn’t a real audience need, an “AI search FAQ” adds little. Structured data can clarify supported information when accurate, but there is no special schema that guarantees a brand mention in generative answers.
Think about reputation and source quality
A generated answer may draw on multiple pages, not only a brand’s own site. Review how trusted sources describe the organisation, and correct factual errors where possible. If a directory or partner page contains outdated details, contact its publisher rather than trying to hide the contradiction with repeated copy on your site.
The best response to a mistaken AI answer depends on the system and the underlying sources. Capture the exact answer and citations, verify the facts, then use the product’s available feedback or correction process. Do not publish a public accusation based on a single unverified output.
Keep expectations modest
Visibility is not the same as demand, trust or business value. A mention can be inaccurate or irrelevant. A cited source may receive no visit. Conversely, a person may learn about a brand in an answer and search for it later. Current reporting may not connect those steps.
Set a question that can be answered with available evidence. For example: “Are our official product facts consistent across our pages?” is actionable. “How often do all AI systems recommend us?” may not be measurable with confidence in 2024. The first kind of work improves the foundation without promising a channel outcome.
A small, repeatable check
If you want to observe a few important queries, choose a fixed set that reflects real customer questions. Record the exact wording, date, country, language, device and product being checked. Capture the full answer and its cited sources, not only the sentence mentioning a brand. Repeat the check later using the same conditions where possible. This will not measure every user’s experience, but it makes your notes comparable and helps separate a recurring pattern from a one-off result. Keep any conclusions modest until you have enough evidence to support them.
Related reading
- AI Overviews: early mistakes and what Google changed
- Author pages and bylines: showing who writes your content
- SearchGPT is announced: should publishers worry?
Pick one important product or service and check how its facts appear across your site and official profiles. What would you want an answer engine to get right first?