Brand visibility in AI is a yes or no, not a position
AI brand visibility is whether an assistant says your name when somebody asks it what to buy. It is also sold as brand visibility in AI search, which means the same thing. That is the whole definition, and the sharpness of it is the point. Search gave you a consolation prize for coming eighth. An AI answer does not. Three or four brands get named and the rest of the market is not mentioned, which from the shopper's side is indistinguishable from not existing.
This makes the measurement unusually clean and unusually brutal. There is no gradual slope from invisible to prominent. For any given question you are either in the answer or you are not, and the answer is short.
It also means the thing you are optimising is not your website's ranking. It is whether the pages an assistant reads happen to contain your name in a form worth repeating. Those are related, because topical authority decides which pages get read at all, but they are not the same thing, and confusing them is why so much effort here produces nothing.
You are not competing for a position. You are competing to be one of the three names in a paragraph, and the paragraph is written from pages the assistant just read.
What actually gets a brand named
Assistants name brands they find stated plainly in sources they trust. Both halves of that sentence matter, and most brands only work on the half they control.
Sources they trust skew heavily toward places you do not own. Community platforms account for the largest single share of what gets cited, because a thread where nine people argue about which product actually worked is exactly the kind of evidence an assistant wants. Established publishers come next. Specialist independent sites follow. A brand's own marketing pages come last, and are frequently skipped entirely when a better source exists.
This ordering is uncomfortable but it is also freeing, because the top tier does not care how old your domain is. A comment carries the platform's authority, not yours.
Stated plainly means quotable. An assistant writing a three-sentence recommendation needs a sentence it can lift. Compare two versions of the same fact. "We pride ourselves on fast, reliable delivery" gives it nothing. "Orders placed before 2pm ship the same day, and we cover the cost of any delivery that misses its window" gives it a claim it can repeat and attribute.
The brands that get named consistently tend to have a specific, checkable thing that is true about them. The brands that do not tend to describe themselves in the same words as everyone else in their category.
Why traffic is the wrong scoreboard
Most businesses judge AI visibility by referral traffic, conclude nothing is happening, and stop. The conclusion is usually wrong and the mechanism is worth understanding, because it will mislead you for as long as you rely on it.
When an assistant recommends three brands, the reader rarely clicks a citation. They read the answer, remember a name, and act on it later. Later means opening a new tab and searching your brand directly, or going straight to your site, or buying on a marketplace. Every one of those paths shows up in your analytics as branded search or direct traffic, with nothing attributing it to the recommendation that caused it.
So the signal you are looking for is a rise in people arriving who already know your name, not a rise in referrals from an assistant. That is a genuinely harder thing to measure, and pretending otherwise is how vendors sell dashboards that count the wrong number confidently.
The workable approach is to measure the thing directly. Ask the questions, record who gets named, repeat on a schedule. That is a real measurement of visibility. Referral traffic is a measurement of clicking, which is a different behaviour that mostly is not happening.
Earning mentions on ground you do not own
Since the most-cited tier is other people's platforms, the fastest route for a brand without much authority is to be genuinely present there. This is also the route most likely to go badly, so it is worth being precise about what works.
Answer the question, mention the product only when it is the answer. Communities are extremely good at spotting someone whose account exists to promote something. The working pattern is a real person who knows the subject, answers things for free, and occasionally says what they make when it is directly relevant. The unworkable pattern is an account that only ever posts links.
Be in threads that already rank. The threads an assistant cites are the ones that surface in search for category questions. Those already exist. Finding the ones that come up for your category and being usefully present in them is far more direct than starting new ones.
Accept the timeline. Perplexity retrieves live and can pick up a new mention within days. ChatGPT tends to move over roughly ten to fourteen weeks. Tracked programmes report a brand going from appearing in one shortlist out of twelve to four out of twelve over about ninety days of consistent participation.
Know the downside. Moderators can and do blacklist a product URL across an entire community, and that is difficult to reverse. This is a slow, genuine-participation channel or it is nothing.
What your own site is still for
None of the above means your own pages are pointless. They do a specific job, and it is worth naming so you invest in them for the right reason.
Your site is the corroboration. When an assistant encounters your brand in a thread and then looks for supporting detail, your pages are what it finds. A store with genuinely useful pages answering real buyer questions confirms the mention. A store with three thin pages and a contact form does not.
Your site is where specific claims live. The checkable facts about what you do, in language plain enough to quote, need to exist somewhere the assistant can read them. That is your product pages, your FAQ pages, and your guides.
Your site is what compounds. Community mentions decay as threads age. Pages that answer questions keep working, and the depth of a real topic cluster is what eventually makes you retrievable on your own account rather than borrowed.
The honest sequencing for a young business is to borrow authority while building your own, rather than choosing one. The store-side guide to AI search covers the on-page half in more depth.
How to measure brand visibility properly
Four rules, each of which exists because the obvious version of it produces a misleading number.
Ask buyer questions, not brand questions. Asking an assistant about your brand tells you what it knows about you. Asking it what to buy tells you whether you exist at the decision. Only the second is visibility.
Repeat, because single answers vary. The same question asked twice can produce different names. One check is an anecdote. Twenty questions asked weekly is a trend.
Check more than one engine. They retrieve differently and disagree constantly. A brand can be a fixture in Perplexity and absent from ChatGPT for the same question, and knowing which is which tells you where the problem is.
Read the answers, not only the tally. Being named dismissively and being recommended both count as present, and only one of them is worth anything. Once a month, read the full text of a handful of answers rather than only the count.
Brand visibility in AI is won mostly on other people's platforms, confirmed on your own, and measured by asking rather than by watching traffic. Get the measurement right first, because the wrong one will point you at the wrong work for months.
The four mistakes that waste the most time
Optimising the page before checking the crawler. If GPTBot or ClaudeBot cannot read your site, none of the rest of it can help. This check takes ten minutes and its outcome is binary.
Writing more pages about a subject nobody asks about. Publishing volume feels like progress and produces nothing when the demand is not there. Measure the questions before writing the answers.
Treating an assistant like a search engine you can rank in. There is no position seven. Effort spent moving from notionally eleventh to notionally eighth is effort spent on a ladder that does not exist.
Buying a dashboard before defining the questions. A tool that tracks the wrong twenty questions produces a confident graph of something that does not matter. The question set is the hard part and it is yours to write, because only you know what your buyers actually ask.
Share of voice, and why your competitors are the benchmark
An absolute visibility number tells you very little on its own. Being named in four answers out of twenty is good or bad entirely depending on what everyone else in your category managed.
The useful measurement is share. Run your question set, record every brand named across every answer, and count. If five brands come up repeatedly and you are one of them, you are in the consideration set. If the same three names appear in nineteen of twenty answers and you appear in one, you know precisely what you are up against and roughly how far away it is.
This reframes the work in a way that tends to be more useful than chasing a score. You are not trying to raise a number. You are trying to join a short list that already exists, and the members of that list are visible to you every time you run the check.
It also tells you which questions to fight for. A question where the same incumbent is named every single time is a hard fight. A question where the answer names a different set of brands each time is one where nothing has settled and a good page can still change the outcome. Unstable answers are opportunities and they are easy to spot once you are recording results rather than glancing at them.
Watch the incumbents rather than only your own line. When a competitor appears in answers they were absent from a month ago, something they did worked, and it is usually findable. A new guide, a thread that took off, a piece of coverage. Copying the mechanism is legitimate and much faster than inventing one.
One caution on reading share of voice. A brand that appears in nineteen answers out of twenty is not necessarily beating you on merit. It may simply have been the obvious name in its category for a decade, which means the pages describing it are everywhere and an assistant cannot avoid it. That is worth knowing because it changes the target. You are rarely going to displace the default name in a category. You are going to become the second or third name in the answer, which is where nearly all of the available ground actually is, and which is a far more achievable thing to aim at.
What to actually do in the first ninety days
In order, because doing these out of order wastes the most time.
Week one: confirm you are readable and write the question set. Check that GPTBot, ClaudeBot, OAI-SearchBot, PerplexityBot and Google-Extended are all allowed. Then write the twenty questions. The questions are the harder half and only you can write them, because only you know what your buyers actually ask before they choose.
Week two: take the baseline. Put every question to two or three assistants and record every brand named. This is the measurement you will be comparing against for the rest of the year, so do it properly once rather than casually four times.
Weeks three to eight: fix the pages that should already be answering. For every question where a competitor is named and you are not, look at what the assistant read. Usually you either have no page on the subject or you have one that says nothing specific. Answer the question directly, near the top, with a checkable claim.
Weeks three to twelve, running in parallel: be present where the answers come from. Find the threads and videos that already surface for your category questions and be genuinely useful in them. This is the slow one and it is why it starts early.
Week twelve: re-run the same twenty questions. Same wording, same engines. Compare against the baseline. Expect movement in Perplexity first and very little in ChatGPT yet, because that is the shape of how these things move.