What service and ecommerce brands are losing while they keep optimising for a search experience that is being replaced. Paid acquisition costs more every year. The organic click is being answered before it happens. And the channel that replaced it is one most marketing plans still have no line item for.
Last reviewed and updated: 19 August 2026. Cited research, tool capabilities and engine behaviour reflect what we could verify as of this date.
01 — What Changed
The AI visibility gap is the distance between how buyers now search and what your marketing is still optimised for. Buyers increasingly ask an engine and get one answer naming two or three brands, instead of getting ten links and working through them. If you are not named in that answer, you were never in the running, and nothing in your analytics will tell you it happened.
That last part is what makes this expensive. There is no impression, no bounce, no lost session. The loss leaves no trace in the reporting you already run, which is exactly why it has gone unmanaged at most brands for two years while budget kept flowing to the channels that can be measured.
Three things are happening at once, and each one is survivable alone.
Paid keeps getting more expensive. Organic clicks are being intercepted by the answer sitting above them. And the new channel that picked up the difference converts better than either, while most brands have no strategy for it because there is no dashboard demanding one.
Auction costs move in one direction over time, and the same budget buys fewer clicks each year than it did. That is survivable while the organic side holds. It stops being survivable when both compress at once, which is what is now happening.
An AI Overview sits above the results a buyer was already going to see and answers the question before the click. The link is still there. Fewer people need it.
Traffic arriving from AI answers converts better than what most brands run, and it is systematically under-counted. Referrer data is often stripped in transit, so much of it lands in Direct and never gets attributed. The result is a channel that looks small in your reporting precisely because it is hard to see.
02 — The Uncomfortable Part
This is the assumption worth testing hardest, because almost every marketing plan quietly depends on it: that being strong in search means being strong in AI answers. The measured data says the relationship is much weaker than that, and that the thing which actually pays is narrower and more specific.
Average organic click-through rate by citation status. Analysis of 53 brands, 5.47 million queries and 2.43 billion impressions, January 2025 to February 2026. Seer Interactive, reported by Search Engine Land.
Those two numbers describe pages sitting on the same result page, for the same query, at similar rank. The only difference is whether the AI Overview cited them. One earns more than double the clicks of the other.
Ranking got you onto the page. It did not get you into the answer, and the gap between those two outcomes is now larger than most of the ranking improvements a brand spends a year chasing.
The cited recommendation inside an answer is not a placement you can currently buy. Ad formats are appearing around these surfaces and will grow, but the name an engine gives when a buyer asks who to use is earned through what the engine reads about you. For a team used to solving distribution problems with budget, that is an unwelcome structural change rather than a campaign to brief.
Citation is winnable, and it is winnable by brands that are not the largest in their category. Engines are not selecting on ad budget or domain age. They are selecting on what the open web says about you and how cleanly it can be read.
That is a rare moment in an acquisition channel, and it closes as categories get organised.
03 — The Diagnosis
Five weighted components out of 100. Presence carries the most because nothing else matters until a buyer hears the name. Accuracy carries the least on weighting and is usually the first thing to fix, because it is the cheapest correction available and the most expensive one to leave alone.
Does the engine name you at all?
The share of the weighted prompt set where the brand is named at least once. This is the floor. Nothing else in the score matters until a buyer hears the name.
A plain identity statement on your own site. The same brand name, category and location stated consistently everywhere else. Presence in the sources the engine already trusts for your category.
When you are named, how much of the answer is yours?
Mentions of the brand as a proportion of all brands appearing in the same answer. Being one of nine names is not the same commercial outcome as being one of three, and a presence-only score hides that difference entirely.
Depth on the specific questions buyers ask, rather than one services or category page. Being described alongside category peers on third-party pages.
Does the engine link back to you as a source?
Whether the answer cites your own domain, as opposed to naming you on the strength of a directory, a marketplace, or somebody else's roundup. This is the component with the clearest measured commercial value attached to it.
A crawlable page that answers the question directly and completely, rather than a page that markets at the reader. Clean structured data. A visible last-updated date that is true.
How are you described once you are named?
Whether the framing is favourable, neutral or negative, judged on the words the engine chooses unprompted. A neutral mention buried mid-list converts very differently to a recommendation.
Review volume and, more than volume, review recency. How your work is characterised by people who are not you.
Is what the engine says about you correct?
Whether services, pricing, location, range and identity are stated correctly. This is the component that shocks people, because the failure mode is rarely a bad review. It is usually an engine confidently describing a product line you discontinued, a price you no longer charge, or a different company altogether.
Disambiguation. State plainly who you are, what you sell, and what you do not sell. Then correct the wrong entity data at its source, rather than on your own site where the engine may never look.
04 — The Bands
A score on its own is a number. The band is the interpretation, written from the buyer side rather than the marketer side, because what matters is what the buyer experiences and not what the dashboard says.
The engine does not know the brand exists. A buyer asking for a recommendation never hears the name, however they phrase the question. Most brands that have never checked are here.
Occasional mentions, usually late in an answer, usually behind three or four competitors named ahead of you.
Named reliably on the softer research questions, and absent on the ones that produce a shortlist.
Named on most prompts including decision intent, but rarely first and rarely cited from your own domain.
The first or only brand named on decision prompts, described accurately, and cited from your own site.
It is a consequence of how an answer gets built. Engines lean on sources that already name brands in a category: roundups, review platforms, directories, industry registers, marketplaces. A brand with an excellent website and no third-party footprint has given the engine nothing to pick up, and reads as though it does not exist. Marketing budget does not fix this on its own, and in several categories the brands spending the most are the least visible.
05 — By Business Model
The same score means different things depending on whether a buyer is choosing a supplier or choosing a product. The failure modes barely overlap, and so neither do the fixes.
A decision prompt returns three to five names. You are either in that set or you are not in the consideration set at all. There is no equivalent of ranking eleventh and still picking up traffic.
Roundups, review platforms, forums and industry registers are what the engine reads when it is asked to recommend a supplier. A brand with a strong site and no third-party footprint has nothing for the engine to pick up.
An engine that describes services you no longer offer, or quotes a price you have not charged in two years, loses you the enquiry silently. Nobody writes in to tell you.
The buyer who would once have asked a colleague now asks an engine first and uses the colleague to confirm. Being absent from the first step means never reaching the second.
Asking for the best product in a category and asking whether your brand is any good are two different questions. They draw on different sources and return different answers. Doing well on one tells you nothing about the other.
The engine names the product and cites the marketplace or aggregator carrying it, not the brand that makes it. You get the sale on a thinner margin, or you get nothing.
AI-referred shoppers convert better and spend more than the rest of your traffic, and a large share of them never appear as AI traffic in your reporting at all. Most media plans are allocating against a number that is structurally too low.
As buying moves toward agents acting on a shopper's behalf, a storefront an engine cannot parse is a storefront it cannot transact with. This stops being a marketing concern and becomes an infrastructure one.
Twelve months earlier the same channel converted 38% worse than everything else. Based on over 1 trillion visits to US retail sites. Adobe Analytics.
06 — Test Your Own Brand
Nothing here needs a tool, a subscription, or us. Substitute your category, your market and your brand name, run each prompt cold on ChatGPT, Google AI Overviews, Gemini and Perplexity, and record what comes back. Most teams find out inside twenty minutes.
| # | Intent | Prompt |
|---|---|---|
| 01 | Awareness | What does a {category} actually do, and do I need one? |
| 02 | Awareness | How much should I expect to pay for {category} in {market}? |
| 03 | Awareness | What goes wrong most often when people buy {category}? |
| 04 | Awareness | Is it worth paying more for {category}, or is cheap fine? |
| 05 | Consideration | What should I look for when choosing a {category} in {market}? |
| 06 | Consideration | What questions should I ask before I commit to a {category}? |
| 07 | Consideration | How do I compare {category} providers in {market}? |
| 08 | Consideration | What separates a good {category} from a bad one? |
| 09 | Consideration | What do people say about {brand}? |
| 10 | Decision | Who is the best {category} in {market}? |
| 11 | Decision | Recommend a {category} for {market}. |
| 12 | Decision | Give me three {category} options in {market} I should contact this week. |
{category}How a buyer says it out loud. Mortgage broker, not credit representative. Waterproof hiking boots, not technical outdoor footwear.
{market}Melbourne, Australia, or the suburb a buyer would actually name. Run more than one if your catchment is local, because the answers diverge sharply.
{brand}Your brand name exactly as it appears on your site. If the engine returns something else entirely, that is your accuracy result arriving early.
Fresh session, logged out where the engine permits it, memory and personalisation off, no follow-up questions. If you run this from an account that has been researching your own brand for a year, the engine will name you and you will conclude you are visible when you are not. This is the single most common way teams get a falsely reassuring result.
Were you named. Who else was named, in what order. What the answer cited. How you were described. Whether that description was actually true. Screenshot each one, because the answer will change and you will want the receipt.
The buyer is asking for a name. This is the prompt that produces the shortlist.
The buyer is building criteria. Being named here shapes who makes the shortlist later.
The buyer is learning the category. Useful to be present, rarely decisive.
Weight the prompts before scoring. A buyer asking who to call is much further down the path than a buyer asking what the category does.
07 — What AI Cites
Presence tells you whether you were named. The source map tells you why, and it is the more useful half. These are the source types worth tracing when you run the prompt set, grouped by kind. They are not ranked: the mix varies enormously by category and by query, and anyone presenting you a universal running order is guessing.
The group most brands have no plan for, and the one carrying the largest single share of citations.
The most-cited domain in ChatGPT by a wide margin. Threads where your category is discussed are frequently what the engine reads before naming anyone.
Consistently among the most-cited domains. Reviews, comparisons and demonstrations get pulled into answers, and the channel does not have to be yours.
Category-specific boards, Facebook groups with public content, Stack-style Q&A. Low volume individually, and they compound.
What an engine treats as settled fact, and what it reads when asked to compare.
Heavily cited for definitional and entity questions. Where an engine goes to confirm what something is.
Third-party articles that name operators or products. A very common route into a decision-prompt answer, and the one brands most consistently ignore.
Rare for mid-sized brands, and close to decisive when it happens.
Sources built to compare and to sell, which is why engines lean on them for shortlists.
Google Reviews, ProductReview.com.au and category equivalents. Drives sentiment far more than presence.
Portals, comparison sites and marketplaces. They frequently absorb the citation that should have been yours.
Yellow Pages, Localsearch, True Local and the category-specific equivalents.
Everything you can edit directly. Necessary for an engine to confirm you, rarely sufficient to get you recommended.
The only source you fully control. It decides whether the engine can read and verify you, and it is not usually what puts your name in the answer.
Cheap to maintain, disproportionately trusted for location, category and hours.
LinkedIn, Instagram, Facebook, TikTok. Better at confirming an entity than introducing one, with LinkedIn the strongest of them for B2B.
Association membership lists and licensing registers. Low volume, high trust.
Analysis of every domain ChatGPT cited across a broad set of US queries, July 2026. Ahrefs Brand Radar.
Almost everything in the community, editorial and commercial groups is somebody else's property. Almost everything in the owned group is yours, and the owned group is where nearly all of the budget, the control and the effort currently goes.
Your own site matters enormously for whether an engine can read and confirm you. It is much less often the thing that puts your name into the answer in the first place. Publishing more pages about yourself does less for AI visibility than being named once, credibly, somewhere the engine already reads.
08 — What To Do
Sequenced by cost and speed rather than by how impressive they sound in a plan. The first two are usually achievable with the team you already have.
Accuracy is the cheapest component to correct and the most damaging to leave alone. Find every wrong fact about your brand sitting in an answer, then correct it at the source that produced it, not on your own site where the engine may never look.
Engines cite pages that resolve a question completely. They rarely cite pages that describe a company. Take the twelve prompts above, find the ones where your category has no good public answer, and write the best one that exists.
Community and user-generated sources carry a large share of citations, and you cannot buy your way in. Being genuinely useful in the places your category is discussed, under your own name and without pretending to be a customer, is slow and it is what the engines are reading.
The comparison articles and review platforms your category already has are a common route into a decision-prompt answer. One credible third-party mention often outperforms a quarter of publishing on your own blog.
Clean structured data, crawlable pages, real last-updated dates, no critical content locked behind scripts. This is unglamorous and it is the part most brands have half-done.
One reading is trivia. Two readings on an unchanged method are evidence. Agree the prompt set, run it cold, record what came back, and repeat it on a schedule rather than whenever someone gets nervous.
09 — Tools & Tracking
A real category of AI visibility platforms now exists. They are worth knowing about, and worth being clear-eyed about. Grouped below by who they are built for, described as at August 2026, because this market is moving quickly enough that any list of features has a short shelf life.
Enough to establish a baseline and watch it move, without a platform commitment or an analyst to run it.
One of the most accessible entry points for tracking brand mentions and links across AI search. Narrower engine coverage than the larger platforms.
Aimed at growth teams. Prompt tracking and competitor comparison across a selection of models, with unlimited seats on its plans.
An add-on to a suite many marketing teams already pay for, which usually makes it the lowest-friction option rather than the most powerful one.
Free, and genuinely useful as a first look. A one-off snapshot rather than monitoring.
Higher prompt volumes, more engines, deeper citation-source analysis, and the governance a larger organisation needs.
Widely treated as the category leader for large teams. Broad engine coverage, high prompt volume, detailed exports, and an action layer beyond reporting.
Strong for large-scale research and citation-source analysis, sitting on top of an existing Ahrefs subscription. Priced accordingly.
Broad engine coverage including Claude, with attention to how content is delivered to automated agents.
Adding AI visibility to existing enterprise SEO suites. Convenient if you already run one, and typically behind the specialists on depth.
It is no longer fair to say these tools only monitor: several ship content suggestions, gap analysis and prompt-level recommendations. The honest gap is further along. A tool can tell you that you are absent from comparison articles in your category. It cannot get you into them, repair wrong entity data at a source you do not own, or decide which of forty recommendations is worth your quarter.
Tools sample their own prompt sets, on their own schedules, with their own scoring. Two tools will give you two different numbers for the same brand in the same week, and neither is wrong. What matters is that whatever you use stays fixed long enough for movement to mean something.
The twelve prompts above, run cold across four engines, will tell you which band you are in for the cost of an afternoon. Buy a tool when you need to watch it continuously or prove movement to someone else, not to find out whether you have a problem.
Engines change their answers week to week and occasionally within a day. A reading is taken at a stated time on a stated method, which is exactly why the same set gets re-run rather than re-invented.
Runs are logged out, memory off, no prior context. Real buyers get answers shaped by their location, history and account. That variation is real, and no published method can honestly average it away.
Being named sits upstream of an enquiry or an order. It is not the same thing, and any consultant treating one as proof of the other is selling you a correlation.
No engine discloses its ranking logic. When a score moves after a change, that is evidence worth acting on. It is not a mechanism, and it will not be described here as one.
Frequently Asked Questions
The questions marketing leads ask once they realise this is both real and measurable.
SEO decides whether you appear in a list of links a buyer then works through. AI visibility decides whether you are named in the answer itself, where there is no list and usually no second page. The two overlap, because engines read the open web, but they fail differently. A brand can rank on page one of Google and be entirely absent when the same buyer asks ChatGPT who they should use.
It is the same territory. We use visibility rather than optimisation because optimisation implies a lever you pull on your own site, and the largest single driver of whether you get named usually sits on somebody else's property. Framing it as an on-site discipline is the most common and most expensive mistake in this category.
It depends on whether you need a baseline or continuous monitoring. Otterly.ai, Peec AI and the Semrush AI Toolkit cover most small and mid-sized needs; Profound, Ahrefs Brand Radar and Scrunch AI are built for larger teams with higher prompt volumes. Several now include optimisation suggestions rather than monitoring alone. None of them will get you named on a third-party source or fix wrong entity data you do not control, which is usually where the actual work sits.
Not in the way you can buy a search ad. Ad formats are appearing in and around AI surfaces, and they will grow, but the cited recommendation inside an answer is not currently a purchasable placement. It is earned through what the engine reads about you. That is uncomfortable for a media budget and it is the actual situation.
Not necessarily, and traffic is the wrong instrument for two separate reasons. Being absent from an AI answer produces no signal at all: no impression, no bounce, no lost session, just an order that went somewhere else. And the AI traffic you do win is routinely misattributed, because referrer data is often stripped in transit and the visit lands in Direct. The channel is under-reported at both ends.
ChatGPT, Google AI Overviews, Gemini and Perplexity cover the great majority of what matters right now. Measure them separately rather than blending them into one number, because they disagree with each other often enough that an average hides the most useful finding. Being named by one engine and invisible in another is common and specifically fixable.
Accuracy corrections can surface within weeks, because they usually involve fixing a source the engine already reads. Presence and share of voice move on a slower cycle, typically a quarter or more, because they depend on third-party sources being published and then picked up. Anyone promising you a citation by a specific date is guessing, since no engine discloses its ranking logic or its refresh schedule.
YouBetterAsk is an AI transformation partner based in Melbourne, working with service and ecommerce brands across Australia and New Zealand on AI visibility, workflow automation and team capability. Not the Djo song, and not Ask.com.
Book a strategy call and ask for an AI Visibility Audit. We run the prompt set across the four engines for your category, trace what each answer cited, and walk you through where you sit and what is worth doing about it.
Prefer to start on your own? The full prompt set and scoring are on this page and you should not need to speak to anyone.