Does Showing Up in AI Answers Actually Bring You Customers?

A founder I know spent four months getting his scheduling product into AI answers. It worked. Ask an assistant which tools a small dental practice should look at and his product came back, named, described accurately, in a decent share of the answers. He had the dashboard to prove it, the line went up and to the right, and he was pleased with himself for about six weeks. His signups over that period were flat. Not slightly up, not noisily up. Flat.

Nothing had gone wrong technically. The pages were good, the citations were real, the score was honest. The problem was upstream of all of it: dentists choosing scheduling software mostly ask the practice next door, or their billing vendor, or a Facebook group of other practice managers. Very few of them open an assistant and ask. He had won a competition his buyers were not attending.

This post is the thing I wish someone had told him in month one. AI search visibility is a real channel with real mechanics, and it also has a specific failure mode that the dashboards are structurally bad at showing you. Here is how to tell which situation you are in, ideally before you spend the four months.

What the score actually measures

Strip the branding off any AI visibility product and the measurement is roughly this: take a set of questions someone might ask an assistant in your category, ask them repeatedly across the major engines, and count the share of answers in which your brand appears. Some also track which sources the answer cited, so you can see whether your page was read even when your name was not used.

That is a genuinely useful measurement, and it is measuring exactly one thing: whether you are found by people who are already asking. It is a conditional probability. Given that someone asks this question of this engine, how often do they meet you.

The number it cannot contain is how many people ask. That figure lives outside the system, it is a property of your buyers rather than of your marketing, and no amount of optimisation moves it. A score of 30% against a question nobody asks is 30% of nothing, and it will look identical on the chart to 30% of something.

The question that decides everything

So before spending on this, answer one question honestly: do your buyers use an AI assistant to make this specific decision?

Not "do they use AI". Almost everyone uses AI now. The question is narrower and much more useful: when they are choosing a vendor in your category, is asking an assistant part of how they do it.

You can answer this in an afternoon without any tooling. Ask ten existing customers how they found you, and then ask whether they have ever put a question like "best X for Y" to an assistant. Ten conversations will tell you more than a quarter of dashboards, and the same conversations are the ones that teach you everything else too, which is the argument in how to get your first customers.

Rough shape of the answer, from what I have seen. Buyers who research software themselves, who are comfortable being early, and who are used to evaluating tools in a browser tab, do ask. Technical founders, indie developers, operations people at software companies, people buying anything they will self-serve. Buyers who choose through a person, a referral, a procurement process, or a local relationship mostly do not, however online they are in the rest of their lives.

When it converts, and when it does not

It converts best where three things are true at once. The buyer researches independently, the category is one where a shortlist is genuinely useful, and the purchase can start without a human in the loop. That is close to a description of self-serve software, and it is why this channel is talked about so much by people selling self-serve software.

It converts worst where the decision is relationship-led or local. If a customer's next step after hearing your name is to ask someone they trust whether you are any good, the assistant did not make the sale, it made an introduction that a referral would have made better.

There is a middle case worth naming, because it is the one people misread. Sometimes you are being cited as a source without being named as a recommendation. The engine reads your page, uses what it says, and credits nobody. That is not failure and it is not success either. It means your content is good enough to inform the answer but your brand is not established enough to be the answer, and the fix is slow: more pages, more specificity, more places where you are the obvious authority on a narrow thing.

The honest limit

Here is the part that gets left off the pitch. Answers often satisfy the person without a click. That is the point of an answer engine, and it means being cited does not reliably produce a visit, so a rising visibility score and flat traffic is a normal combination rather than a broken one.

Attribution is also genuinely poor. Referral data from assistants is patchy, so a lot of what they do send arrives looking like direct traffic. This cuts both ways: it means some of the channel's real value is invisible, and it also means the channel is unusually easy to sell, because a vendor can always explain a flat number by saying the good part is unmeasurable. Be suspicious of that explanation, including when it comes from us.

And the honest version of the whole thing: for most businesses today, AI search is a supporting channel, not a primary one. It is cheap to do alongside content you were writing anyway, it compounds, and it is winnable earlier than Google because the cited pool is small. It is not a substitute for knowing where your buyers actually are, which no channel can tell you in advance, as best marketing channels for startups works through.

What to measure instead of the score

Watch the number that has a customer at the end of it. Signups, conversations, demo requests, the things that pay you.

Three practical instruments. Add a "how did you hear about us" field to signup and actually read it, because self-reported attribution is unfashionable and still the single most useful signal a small company has. Watch direct and branded search volume, since an assistant that names you often produces a person who types your name rather than clicking a link. And note whether the language your prospects use is starting to match the language your pages use, which is the earliest sign that answers are shaping how people arrive.

Give it a fair clock before you judge it. Citations start appearing within weeks of publishing, but the brand recognition that turns a citation into a visit builds over months, so a four-week read is too early to conclude anything and a six-month read with nothing behind it is conclusive. Pick the window in advance and write it down, because the failure mode here is not impatience, it is the opposite: a score that rises pleasantly forever while you postpone the question of whether anything arrived.

If those stay flat across a few months while the visibility score climbs, believe them and not the score. The tactics for improving the score itself are a separate question, covered in how to get recommended by ChatGPT, and the build-versus-buy view is in AI SEO vs an SEO agency.

Where Revnu fits

Revnu runs this work as one lane among several, which is the only framing we think is honest about it. It writes the pages that earn citations, tracks where you appear across engines, and reports what moved. It also runs outbound, ads and the rest, in one place with one memory, so when a channel is not producing you can see it against the others rather than in isolation, and move the effort. Every draft waits for your approval. The reason we build it that way is the failure in the opening story: a single-channel tool has no way to tell you that its own channel is the wrong one, because the only number it can see is its own.

Where this leaves you

Being in AI answers is worth having and it is not a growth strategy by itself. The score tells you whether you are found by people already asking, which is genuinely useful and is not the same as demand. So before you spend, ask ten customers whether they would ever put your category question to an assistant. If the answer is mostly no, do this cheaply alongside your content and expect a slow compounding benefit rather than a pipeline. If the answer is mostly yes, you have found a channel worth real effort. Either way, judge it on signups rather than on share of voice, and be willing to conclude that you won a competition your buyers were not attending.