Your next customer may decide before they visit you

Marketing measurement was built on a lucky coincidence. For twenty years, the places where buyers formed their impressions of you (search results, your website, your ads, the review pages) were also places someone could observe. Rankings were public. Search volume was purchasable data. Every visit left a session behind. Influence and observability travelled together, so completely that an entire discipline grew up assuming they were the same thing.

They were never the same thing. And AI assistants are now pulling them apart.

When a buyer asks an assistant to compare options in your category, the response is not a list of links to click through. It is a synthesised answer (a summary, a comparison, often a recommendation) assembled from many sources and delivered whole, before any click happens. The work a buyer used to do across ten open tabs, in full view of ten analytics scripts, now happens inside one response. Shaping a first impression no longer requires a visit at all.

That much is not a forecast. You can verify it yourself in the next sixty seconds.

The sixty-second test

Open any major AI assistant and ask it to recommend three vendors in a category you know deeply (your own, ideally). Watch what it does.

It will almost certainly answer. Not with a hedge, not with a list of considerations, but with named options, specific comparisons, and, read closely, a lean. And it will do this without asking you a single clarifying question, which is the tell. A good salesperson would ask who you are, what you're solving, what you've tried. The assistant has already decided it knows enough. It has formed a point of view and delivered it, and if that answer had gone to a real buyer instead of you, no one from your company would have been in the room.

Be precise about what this test proves. It proves the mechanism is live: these systems really do compress discovery and evaluation into a single synthesised verdict, today, in your category. It does not prove how many buying decisions that verdict actually moves. That second number doesn't exist yet: not for us, not for the platforms, not for anyone selling you an answer to it.

What you can infer, and what nobody knows

Between what's demonstrable and what's unknown sits a band of reasonable inference, and it's worth being explicit about which claims live where.

You can reasonably infer that some buyers act on these answers, because acting on them is what answers are for. A response that names three vendors and leans towards one is doing the job a shortlist used to do; it would be strange if none of the people asking treated it that way. You can also reasonably infer the influence is largest exactly where research effort used to be largest: considered purchases, unfamiliar categories, anywhere a buyer once opened those ten tabs.

What nobody knows: how many buyers, in which categories, with what conversion effect. Anyone quoting you a confident market-wide figure today is ahead of the evidence.

And one further uncertainty, less discussed: there may be no single "answer" to see. Ask the same buying question twice and you may get different shortlists. Your brand's representation inside these systems may not be a position, like a search ranking; it may be a distribution. Which would mean the question "what does AI say about us?" has no one answer, only a shape. No one has published a good measurement of how stable these shortlists are. That is a genuinely open question, and it should unsettle you more than any single bad answer would.

Why your dashboards will be the last to know

The problem here is structural: not a gap in your analytics configuration, but a gap in what analytics can be.

Everything you measure begins at the click. Session, source, funnel, attribution: all of it starts when someone arrives. That was an acceptable blind spot when the upstream world was observable by other means: you could see the rankings everyone saw, buy the search-volume data, read the same reviews your buyer read. The upstream was public, so your measurement could afford to start downstream.

An AI-mediated first impression is different in kind, not degree. It is private (one buyer, one conversation) and personalised by how the question was phrased. And it is ephemeral: nobody else will ever see the exact answer your buyer saw. There is no rank-tracker for a private conversation. The first impression has become a private event.

Now follow that through to your funnel. The buyer whose synthesised comparison omitted you never clicks, never becomes a session, never enters the pipeline, never shows up as a lost deal with a reason code. They don't appear in your data as a loss. They don't appear at all. Which means the effect, when it grows large enough to matter, will not announce itself as a cause. It will look like unexplained drift (branded search softening, inbound thinning, win rates holding steady while volume quietly sags) with nothing in any dashboard to attribute it to. The first symptom will be silence.

And the risk hiding inside that silence isn't visibility. It's interpretation. An assistant that misdescribes your positioning, repeats a claim that stopped being true two pricing models ago, or compares you against the wrong peer set is actively shaping perception; there is currently no standard process, at almost any company, for catching it, let alone correcting it, before the perception hardens.

What to do while it's cheap

None of this argues for abandoning the journey you can see. The homepage, the product page, checkout: they still convert every buyer who arrives, and swinging budget wholesale towards a stage nobody can yet measure would be trading a working discipline for a fashionable guess. The old journey isn't dead. It has grown a new first stage, and the task is to extend your attention upstream without dismantling what works downstream.

Four moves, in ascending order of effort:

Assign the question. Today, at most companies, nobody owns "what do AI assistants say about us." Not brand, not SEO, not comms: nobody is assigned to know. Give the question a name and an hour a month. It costs almost nothing, and at most companies it is not currently anyone's job.

Run the audit. Ask the major assistants to describe your company, compare it to competitors, and recommend options in your category, phrased the way a buyer would phrase it, not the way you would. Log what comes back: missing facts, stale claims, absent comparisons, wrong peer sets.

Treat what you find as debt. Not as a marketing nice-to-have, but as content and structured-data debt. Like technical debt, it accrued without anyone deciding to take it on, it compounds quietly, and it gets more expensive to fix the longer the wrong version of you circulates.

Start a baseline. An emerging diagnostic is worth knowing about here: AI Share of Recommendation: how often your brand appears in AI-generated answers to category-level buying prompts, relative to your competitors. It is the natural analogue of share of search, and it is not settled science: the methodology isn't standardised, and anyone selling it as a precise index is overselling. Measure it anyway. An early baseline won't be precise. It will show you direction, and it will mean someone at your company is finally looking.

A baseline in one afternoon. No tooling required. Pick one category you know well. Write down, in advance, your five real competitors and five facts about your company that are currently true. Then ask three assistants the same three buying questions (describe, compare, recommend) five times each, in fresh conversations, and log by hand: who was named, who was recommended first, and which of your five facts came back wrong or stale. It is deliberately small. It will still tell you two things almost no company knows about itself: roughly how often you appear, and whether the answers even agree with each other.

The discount ends

Why act on a shift you can't yet fully measure? Because of the order in which the proof will arrive.

By the time AI-mediated discovery is unambiguous in revenue data, it will be unambiguous in everyone's revenue data: the same quarter, the same board decks, the same scramble. The corrective work, meanwhile, has lead time: corrected facts take time to propagate, missing structure takes time to build, and an absence has to be noticed before it can be repaired at all. Advantage lives in the gap between a shift being demonstrable at the mechanism level and its being priced in at the market level. That gap is open now.

Catching up is still cheap. It stays cheap only while your presence in those private first impressions remains a blind spot no one has priced, including your competitors. The moment it becomes a number on everyone's dashboard, the discount ends.

For twenty years you optimised every moment of the journey you could see, and it worked. The next gap opens in the moments you can't. The companies that close it first won't be the ones with the best answer to what AI is saying about them; for now, they'll simply be the ones who know.

Beyond the Click: one AI-mediated customer journey shift, one brand implication, and one metric to track, every week. Subscribe to get the next issue.