Your branded search is falling: how to tell if the AI is answering for you
Of all the metrics that can fall in an account, branded search — the searches that include your client's name — was the sacred one. Generic traffic rises and falls with the algorithm, but if someone types "Robles Dental Clinic reviews", that click was yours no matter what: nobody competes for your client's brand better than your client. So when branded searches start dropping without the business itself dropping, it's worth not looking the other way: either the brand is losing interest, or someone is answering those questions before they reach Google.
That "someone", more and more often, is an AI. And the problem has an awkward property: it happens outside your tools. If a prospect asks ChatGPT "is Robles Clinic any good?" and takes the answer at face value, there's no impression in Search Console, no session in Analytics, no trace anywhere you normally look. Only the gap it leaves behind.
The signs in the data
None of these signs proves anything on its own; together they draw the pattern. Before pointing the finger at AI, rule out the obvious: seasonality, less media spend (branded search lives partly off advertising and awareness), a reputation crisis or a half-finished rebrand.
With that ruled out, the signs that point to AI:
- Branded searches fall, but the business doesn't. Same sales, same leads, same calls — fewer people searching the name on Google. The demand exists; it has changed door. It's the branded version of the general phenomenon: zero-click searches have gone from 56% to 69% in a year (data cited on stucom.com), and people increasingly resolve questions without setting foot on a website.
- What drops most are the branded searches "with a surname". "Robles" holds up, but "robles clinic reviews", "robles clinic prices", "robles clinic or sonrisalud clinic" collapse. Those research queries are exactly the ones a user now dispatches in a conversation with ChatGPT: the AI gives them the reviews, the price range and the comparison in a single message.
- Direct traffic and visits "out of nowhere" hold or rise. Users who reach the site by typing the URL or from AI sources, without going through branded search. They got to know and assess the client somewhere else; they only come to seal the deal.
- AI Overviews already answer the brand question. Search "[brand] reviews" on Google: if an AI-generated summary appears above the results, the user no longer needs the click. Recall the general figure: where there are AI Overviews, the average organic CTR falls 61% in Spain (ismajimenez.com). That your client is the first result below the summary is little comfort.
How to check it: the 30-minute audit
You confirm the suspicion by asking the AIs the same things a prospect would ask. The script:
Step 1 — The brand questions. In ChatGPT, Gemini and Perplexity (clean sessions, no history to contaminate them), fire the four questions that make up a decision: "what's [brand] like?", "is [brand] reliable?", "[brand] versus [main competitor], which is better?" and "how much does [service] cost at [brand]?".
Step 2 — Note three things about each answer. Does it answer confidently or say it has no information? Is what it says correct and current (prices, services, location)? Who else does it mention — does the competitor show up in the answer about your client?
Step 3 — The category questions. "Best [service] in [city]", "recommend me a [type of business] in [area]". Here you measure whether the AI offers up your client when nobody names them. Don't be surprised if it doesn't: in the study we ran on 9,865 Spanish SMEs across 30 sectors, 91% appeared in only 1 of the 4 main AIs (full study).
Step 4 — Repeat in two weeks before concluding. AI answers are volatile and a single snapshot misleads; two separate snapshots already sketch something. Why serious measurement works this way — sampling, trends, not loose screenshots — we explain in can AI visibility be measured?.
The audit yields three possible diagnoses, each with its own severity: the AI can't answer about your client (it loses the research queries, but it's recoverable), the AI answers well (the branded-search drop is just the lost click — annoying, not serious), or the AI answers wrong or with errors: old prices, services that no longer exist, mix-ups with another company. This last one is the urgent one, and it has its own protocol in what to do when ChatGPT says something false about your client.
What to do depending on what you find
If the AI doesn't know or answers partially: give it something to read. The AIs compose their answers from public sources: the client's website (does it actually answer "prices", "reviews", "opening hours", or does it hide them behind a form?), their Google Business listing, sector directories, reviews and third-party mentions. An honest pricing page and a well-built FAQ work harder for the AI than ten filler posts. The full process is in the first AI visibility audit, step by step.
If the AI answers with errors: prioritise correcting the sources that feed the error (an outdated Google listing, a directory with old data, a 2023 pricing page) and document the before and after.
If the AI answers well: congratulations, and now protect it. Today's answer doesn't guarantee the one in three months, and the competitor who runs this same audit will want that spot. This is where the one-off check turns into monitoring: we do it with Surfeo, which asks the 4 AIs about each client's brand and category every week and alerts you when an answer changes — because branded search used to be watched in Search Console, but its replacement you have to go and look for.
In any of the three cases, the finding is commercial gold: few things move a client more than seeing with their own eyes what the AI says (or doesn't say) about their brand.
Frequently asked questions
Where exactly do I see branded search?
In Search Console: Performance → query filter → contains "[brand]" (add variants with common typos). Compare the last 12 months with the previous 12. Always separate it from generic traffic: mixed together, the brand drop gets buried in the noise.
Couldn't the branded-search drop simply be a loss of awareness?
It could be, and that's why step one is to cross-reference it with the business. If sales, leads and calls fall at the same rate, the problem is the brand, not the channel. If the business holds while branded search drops, the demand is still there and only the route has changed: that's where you investigate the AI.
Can I stop the AIs from talking about my client?
In practice, no — and it almost never serves you. Blocking AI crawlers in robots.txt doesn't erase what the AI already knows from third parties: it only takes away your voice in the conversation and lets reviews, directories and competitors fill it. The sensible strategy is the opposite: make sure the sources the AI consults tell the correct version.
How often do I repeat the check?
Manually, once a quarter per client is the decent minimum; AI answers change with every model update. If you manage several accounts, the hours maths gets absurd fast — that's what weekly monitoring tools are for, turning it into a report that arrives on its own.
Suspect the AI is already answering for one of your clients? Settle it in five minutes: take the free AI visibility test and see what each AI says about their brand.
