TL;DR
📉 The click was a proxy, not the prize: Ahrefs measured a lower clickthrough rate for the top-ranking page when an AI Overview sits above it, while 51% of software buyers now begin research inside an AI chatbot.
🪜 Being named still counts: one-third of buyers purchased from a vendor they had never heard of before.
🧪 Rewrites are not the lever: C-SEO Bench found only 3 of 54 tested conditions produced significant citation-rank gains, while the KDD 2024 GEO study raised visibility by adding statistics, quotations, and citations.
📚 Most of the answer is not yours: 57% of AI citations point to sources brands do not control.
🎯 Get on the list before the call: buyers consider roughly 5 vendors and 3.8 are already on the list before they contact anyone.
🔭 Say what you cannot see: the IAB splits AI visibility into 4 separate rungs - Presence, Prominence, Portrayal, and Persuasion - rather than one number.
A traffic dashboard cannot tell the difference between demand disappearing and a click disappearing. Those are separate events with separate causes, and for most startups right now only the second one is happening.
Ahrefs measured a lower clickthrough rate for the top-ranking page when an AI Overview sits above it, and Ahrefs found 38% of those overview citations rank in the organic top 10 as of March 2026, against roughly 76% in July 2025, so the named sources are increasingly not the pages ranking beneath them. The visit was never the thing you wanted. Being considered was.
So the operating answer is short. When the answer resolves the question, presence inside that answer becomes the earlier and more decisive event, and the click becomes optional.
6sense mapped where B2B sites are losing traffic to LLMs and found the loss concentrated in exactly the pages that used to answer early research questions. Gartner projected search engine volume would fall 25% by 2026 because of chatbots and virtual agents, which is a projection and not a measurement, and Similarweb's tracking of ChatGPT referral traffic shows the replacement channel is real but much smaller than what it displaced. Your job is to be in the answer, and then to report what you can and cannot see about it.
What are you actually losing when the click disappears?
You are losing an observation, not necessarily a buyer. The click was a measurement artifact that happened to be free: someone considered you, and the browser wrote it down. When an assistant answers in place, the consideration still happens and the log entry does not.
That distinction decides where the budget goes. If demand has moved, you have a product and positioning problem. If only the record has moved, you have a measurement problem plus a presence problem, and those are cheaper to fix. Profound's analysis of a 7.5 million-conversation sample found commercial conversations in ChatGPT more than doubled in a year, which is evidence that buying questions are being asked, just not on your property.
Commercial conversation volume is not a purchase queue. A separate corpus of 8,133 multi-turn human-LLM conversations found exploration markers in 50.3% of commercial conversations and 44.8% of PRISM conversations, so a large share of commercial-looking volume is research, and the safe reading is that early research moved into a system you do not instrument.
The competitive shape changes too. G2 found AI chatbots are now the single largest influence on B2B shortlists, 51% of software buyers now begin research inside an AI chatbot, and one-third of buyers purchased from a vendor they had never heard of before. Read the last one as a doorway: an unknown vendor can enter the consideration set at the answer layer, with no brand recognition behind it. That is the opportunity hiding inside the traffic decline.
Which questions still deserve a page from a small team?
Fewer than your content calendar assumes, and they are the questions a buyer asks immediately before choosing. Broad informational pages are the exact category the answer now absorbs, so publishing more of them buys you less every quarter.
Pick the questions by working backwards from the decision. What does someone compare you against, what disqualifies a vendor, what does implementation cost in time, what breaks. 6sense found buyers consider roughly five vendors and that 3.8 of them are already on the list before the buyer contacts anyone, so the pages worth writing are the ones that decide list membership. 6sense's 2025 B2B Buyer Experience Report describes the same compressed window from the buyer's side.
Then resist the tactical shortcuts, because the research does not support them. C-SEO Bench, published at NeurIPS 2025, found only 3 of 54 unilateral conditions produced statistically significant citation-rank gains. Its code and setup, which tested conversational-SEO methods across two tasks and six domains, are public, and the finding is blunt: rewriting a page in an AI-friendly voice is mostly a no-op. Choosing better questions is the work that pays.
One counterexample to keep honest: the GEO study at KDD 2024 found that adding statistics, quotations, and citations raised visibility in its benchmark, while phrasing changes did not. The methods that worked added evidence. The methods that failed added adjectives.
What kind of evidence survives extraction?
Evidence that can be lifted out of your page and still be true. An assistant does not quote your argument, it quotes a sentence, so a claim that depends on three surrounding paragraphs to be accurate will either be dropped or mangled.
Start with the mechanics, because nothing downstream matters if the page cannot be read. Vercel's crawler research with MERJ documented that OpenAI's and Anthropic's crawlers do not execute JavaScript, while Google's indexing path applies the same Web Rendering Service capabilities that power classic search. Cloudflare's agent-readiness study across the 200,000 most visited domains measured how much of the web an agent can actually read. Fetch your own pages with scripting disabled before you write another word.
Then make each claim self-contained: the number, the source, the date, the conditions it holds under. An AirOps analysis of 548,534 pages mapped which page traits track with being pulled into answers.
Classic retrievability still matters, with a caveat about which surface you are talking about. Seer's sample of 500 SearchGPT citations found 87% matched Bing's top results, so on that engine ranking sits upstream of citation. On Google AI Overviews the Authoritas analysis above shows a looser relationship. The two surfaces do not behave the same way, and a single retrievability number will hide the difference.
Skip the persuasion reflexes. Scarcity and exclusivity framing measurably reduces how often a model recommends a product, in a setup built on 10 fictitious products so no real brand equity could carry the result. Copy instincts that work on a landing page can cost you inside an answer. Much of what decides that answer sits off your domain: 57% of AI citations point to sources brands do not control.
How do you measure presence when sessions stop reporting it?
You measure appearance rates on a fixed set of questions, repeated over time, and you accept that the unit of measurement is a distribution. A single answer tells you almost nothing.
SparkToro found AI engines are highly inconsistent when recommending brands, with the same prompt producing different vendor lists across runs, and Search Engine Land's write-up of a separate study reported the same instability. A 2026 variance-components study found run-to-run noise large enough to swamp real differences in small samples, and work on quantifying uncertainty in AI visibility makes the case for reporting confidence intervals.
Use a framework that separates the rungs instead of averaging them. The IAB's AI-era visibility guidance breaks the problem into Presence, Prominence, Portrayal, and Persuasion, and its accompanying guidelines sort the work into 4 groups for brands. A mention is not a citation, a citation is not a recommendation, and a recommendation is not a shortlist slot. Collapsing those four into one score destroys the only signal you had.
Keep the site-side numbers you still trust, but demote them. GA4's key event rate is sessions with a key event divided by total sessions, which stays meaningful for the people who do arrive and says nothing about the people who never needed to. Report both, and let the second silence stand as silence.
Semrush's AI Visibility Index, built on 126 million AI search prompts, is one external reference point for category-level movement. It is a composite, so use it to see where your category is heading, and keep your own reading on each rung separate.
What do you do about the part of the funnel you cannot see?
State it as an unobservable region with a stated size, and stop filling it with invented attribution. The worst outcome of the click decline is not lost traffic, it is a board deck that assigns credit to whichever channel still emits a parameter.
The honest reporting move is a named line item: early research now happens inside assistants we do not instrument; here is measured presence on the questions that matter, the run-to-run variance on it, and the direct and self-reported attribution we treat as its downstream shadow. That is defensible. Back-fitting pipeline to a visibility score is not.
Two supporting facts make the case internally. 6sense found 58% of buyers say they engaged sellers sooner because of AI, which means the pipeline can look healthier while the top of the funnel looks emptier. And a 2026 analysis of brand dynamics in LLM recommendation systems shows prior brand associations are sticky and unevenly spread, so presence compounds slowly.
Ask sales the question no dashboard answers: what did the buyer already believe on the first call, and where did they get it.
How does Trovance help when the answer replaces the visit?
Trovance treats presence in the answer as the measured event. You define the buyer questions your market actually asks, and it runs them repeatedly across AI engines as tracked questions, preserving each answer run as a snapshot with its full context: who was mentioned, who was cited, who was recommended, and which sources carried the answer. Answer coverage over time is the reading.
A fixed question set is a proxy. Real buyers arrive mid-conversation with context you did not write, so treat the panel as a repeatable instrument and not a transcript of demand. Its value is that it holds still while the engines move.
That record is what makes the click decline interpretable. Because every snapshot keeps its citations, you can see whether a competitor won on their own pages or on a third-party source you never touched, and whether your absence is a retrieval failure or an evidence failure. Your Brand Core holds the claims you are entitled to make and the proof behind each one, so recommended actions name the missing asset and the evidence it needs. Drafts are produced from approved claims, and a person reviews everything before it publishes.
What Trovance will not promise is a citation, a ranking, or a recommendation. The engines are probabilistic, the variance is documented above, and any product that guarantees an outcome inside someone else's model is selling against the evidence. There is no single visibility score here either, because collapsing presence, citation, and recommendation into one number hides the rung you are losing. We are building toward tighter linkage between answer presence and downstream pipeline, and we will say plainly which parts of that chain remain unobservable.
What it does close is the loop. After an asset ships, the next analysis cycle reruns the same questions and compares the new answers against the preserved ones, so you can verify whether the answer moved instead of assuming a traffic chart explained it.
What should you do this week?
Write down ten questions a real buyer asks in the two weeks before choosing in your category. Run each one at least ten times, across at least two engines, on different days, and record mentions, citations, recommendations, and shortlist slots separately. That baseline takes an afternoon, and it gives next quarter's numbers a fixed reference point to move against.
Then fix retrieval, which is mechanical and fast: fetch your key pages with scripting off and see what an engine reads. Put one defensible number, with its source and date, on each page that answers a decision question. Check which third-party sources your engines cite in your category. Finally, add the unobservable-funnel line to your reporting before someone else fills it with a made-up model.
Do not expect this to move in days. In our experience the order of latency runs retrieval first, then evidence, then earned third-party coverage. Nobody has published reliable intervals for any of the three, so treat that as an order of operations. If you want the baseline without the manual runs, start a free Trovance analysis and watch your buyer questions.
Measure what replaced the click
How to attribute AI traffic in GA4 - what the analytics you already have can and cannot show.
How to measure AI search visibility without one score - why the rungs have to stay separate.
Do AI search visitors convert - what the smaller referral stream is worth per session.
Does AI search visibility drive leads or revenue - the evidence for and against the pipeline claim.
Earn presence inside the answer
What gets cited by AI - the page traits that track with being pulled into answers.
How to earn third-party AI citations - the majority of the answer sits off your domain.
AI agents need evidence, not more content - why volume stopped working first.
A visibility gap is not a content brief - when the correct response to a gap is no page at all.
FAQs
How can startups win when searches end without clicks?
By competing for presence inside the answer instead of the visit. Pick the ten questions buyers ask right before choosing, put extractable evidence on the pages that answer them, and measure appearance rates across repeated runs. G2 found AI chatbots are now the single largest influence on B2B shortlists.
Does a drop in organic traffic mean demand is falling?
Not by itself. A traffic chart measures clicks, and clicks were always a proxy for being considered. Ahrefs measured a lower clickthrough rate for the top-ranking page when an AI Overview appears, which is the proxy disappearing rather than the demand. Falling demand shows up in pipeline and win rates instead.
Should a small team publish more content to offset zero-click search?
Usually no. C-SEO Bench found only 3 of 54 tested conditions produced significant citation-rank gains, so volume and phrasing tricks rarely move anything. The KDD 2024 GEO study points at evidence instead: adding statistics, quotations, and citations raised visibility in its benchmark, while phrasing changes did not.
How do startups measure AI presence without session data?
Run a fixed set of buyer questions repeatedly and record mentions, citations, and recommendations separately. The IAB's AI-era guidance separates Presence, Prominence, Portrayal, and Persuasion for exactly this reason. Report a distribution across runs, because a 2026 variance study found noise large enough to swamp real differences in small samples.
Is being cited without a click worth anything?
Yes, when the citation reaches a buyer forming a shortlist. G2 research found one-third of buyers purchased from a vendor they had never heard of before, which means an unknown vendor can enter the set at the answer layer. 6sense found buyers consider roughly five vendors, with 3.8 already on the list before contact.
What should we tell leadership about the traffic we cannot see?
State it as a named unobservable region with a stated size. Report measured presence on tracked questions, the run-to-run variance on that measurement, and direct plus self-reported attribution as the downstream shadow. 6sense found 58% of buyers say AI led them to engage sellers sooner, so pipeline can improve while the funnel top empties.
How long does it take to show up in AI answers?
It depends on the failure mode. In our experience retrieval fixes surface first, evidence changes follow, and earned third-party coverage lands last. Nobody has published reliable intervals, so treat that as an order of operations. No honest vendor guarantees a citation or a recommendation on any timeline; measurement variance alone forbids it.
