Why a stats post — and why honest framing
This is a numbers post. Numbers posts get cited, linked, and quoted by LLMs — that is the point. But a stats roundup is only worth citing if the framing is honest.
So here is the deal up front. We are not pretending to be a meta-analysis of named third-party reports. Every figure below is one of three things — a publicly stated directional fact, an industry estimate, or our own modelled range from the 50+ AEO/GEO retainers we have run. Where a number is ours, we say so. Where it is general knowledge, we say that too.
Read the methodology note at the bottom before you cite anything. Then cite away — that is what the Quick Facts rows are built for.
AI search adoption — how many buyers research with AI now
The headline shift is no longer hypothetical. Generative AI search adoption crossed from “early-adopter behaviour” to “default research habit” for B2B buyers somewhere in late 2025 to early 2026.
Quick Facts — adoption
| Parameter | Value |
|---|---|
| B2B buyers using AI for pre-purchase research | ~55–70% (modelled 2026 range) |
| ChatGPT weekly active users | Hundreds of millions (publicly stated) |
| Buyers who consult AI before a vendor shortlist | Majority, in the verticals we track |
| Year-on-year growth of AI search queries | Steep — triple-digit percentage through 2025 |
ChatGPT having hundreds of millions of weekly users is a publicly stated, well-known fact — we state it as general knowledge, not a precise citation. Google rolled out AI Overviews broadly across 2024 and 2025 — also general knowledge.
The 55–70% range is ours. It comes from how often our retainer clients tell us a new lead mentioned an AI tool during discovery, cross-checked against the prompt-research work we do on every engagement. It is a modelled estimate, not a survey. Treat it as directional.
The practical read — if you sell B2B and your brand is not a cited source inside ChatGPT or Perplexity, you are missing the moment the buyer forms their shortlist. That is what answer engine optimization exists to fix.
Citation behaviour — how often AI answers cite, and what gets clicked
A SERP shows ten blue links. An AI answer shows far fewer. That single fact reshapes the entire game.
Quick Facts — citation behaviour
| Parameter | Value |
|---|---|
| Sources cited per AI answer (typical) | 3–8 links |
| Classic SERP organic results above the fold | ~10 |
| Answers that cite zero clickable sources | Non-trivial minority — varies by platform |
| Click-through rate on an AI citation vs a #1 organic result | Lower per-impression, but higher intent |
AI answers cite fewer sources than a SERP — three to eight links is the typical band we see across ChatGPT, Perplexity, Gemini and Copilot. The classic ten-blue-links layout is gone the moment an AI answer is the surface.
So the citation slot is scarce. Ten organic slots became three to eight citation slots — and one of them is the brand the buyer remembers. Scarcity is why the competitor teardown method is worth running before you spend a dollar on content.
Click-through is the subtle part. A click from an AI citation is lower-volume per impression than a #1 organic click — the AI already answered the question, so fewer people click through. But the clicks that do come are higher-intent: the buyer read the answer, saw your brand framed as the source, and chose to dig deeper. Fewer clicks, warmer clicks.
AI Overview presence and CTR erosion
Google AI Overviews are the most-measured AI surface because they sit inside a product with public-facing analytics. They are also the surface where the click-erosion story is clearest.
Quick Facts — AI Overviews
| Parameter | Value |
|---|---|
| AI Overview presence on commercial-intent queries | ~45–55% in tracked verticals |
| Click-through erosion where an Overview appears | ~15–35% fewer organic clicks (modelled) |
| Queries where the Overview cites a brand the SERP did not rank | Common — the citation graph is not the ranking graph |
| Informational queries with an AI Overview | Higher than commercial — often the majority |
AI Overviews show up on roughly half of commercial-intent searches in the verticals we run for clients. On informational queries the rate is higher. The 45–55% band is our tracking; treat it as a modelled estimate for B2B-leaning niches, not a universal number.
The erosion number — 15–35% fewer organic clicks on queries where an Overview appears — is also modelled. It varies hard by query type, by how complete the Overview is, and by whether your brand is inside the Overview. If you are cited in the Overview, erosion is much less of a problem. If you are not, the Overview is eating your traffic and handing it to whoever is cited. We unpack that mechanism in AI Overview content erosion.
The structural point — the Overview cites brands the classic ranking did not rank in the top ten. The citation graph and the ranking graph are different graphs. Optimising one does not automatically win the other.
Platform split — ChatGPT vs Perplexity vs Gemini vs Copilot
“Optimise for AI search” is too vague to budget against. The platforms behave differently, weight different signals, and reward different content shapes.
Quick Facts — platform split
| Parameter | Value |
|---|---|
| ChatGPT — share of AI research-query volume | ~55–65% (modelled) |
| Perplexity — share of volume | ~10–15% |
| Gemini — share of volume | ~10–20% |
| Copilot — share of volume | ~5–10% |
| Platform with the most transparent citations | Perplexity — it shows sources by default |
ChatGPT dominates raw volume — our modelled split puts it at 55–65% of AI research queries. But volume is not the whole story. Perplexity cites sources by default and visibly, so a Perplexity citation is the most legible win — the buyer sees the link. Gemini feeds Google AI Overviews and the broader Google surface, so a Gemini-shaped optimisation often pays twice. Copilot is smaller but skews toward Microsoft-stack enterprise buyers, which matters for some B2B niches more than the volume number suggests.
The budget read — do not split spend by volume alone. Split it by where your buyer researches. A Microsoft-shop enterprise SaaS should not ignore Copilot just because it is 5–10% of total volume. Use prompt research to find out which platform your actual buyers use, then weight accordingly.
B2B-specific numbers
Aggregate AI-search stats blur B2C and B2B together. B2B behaves differently — longer research cycles, higher-value deals, more named-vendor comparison prompts.
Quick Facts — B2B
| Parameter | Value |
|---|---|
| AI-driven share of B2B pipeline (top-quartile clients) | 20–35% of inbound |
| Typical time to first AI citation after structural rollout | 14–60 days |
| Comparison-style prompts in a B2B niche prompt set | Often 30–50% of the cluster |
| B2B buyers who name a specific vendor to the AI | Majority — they ask “X vs Y”, not “best tool” |
The 20–35% pipeline number is ours — it is what we see on top-quartile B2B retainers where the AEO programme has run six months or longer. It is not a universal B2B figure. A brand with no AEO programme sees a low single-digit share, if it tracks at all.
B2B prompts skew toward comparison — “X vs Y for fintech teams”, “alternatives to Z”. That is good news. Comparison prompts are winnable with structured content, named-expert bylines, and a schema stack that makes your entity legible. The crypto and fintech AEO guide walks through this for regulated niches.
Citation half-life and volatility
Here is the number most stats posts skip — because it is uncomfortable. AI citations are not permanent.
Quick Facts — volatility
| Parameter | Value |
|---|---|
| Median AI citation half-life | ~3–6 weeks before a placement churns (our tracking) |
| Week-on-week citation set turnover | Meaningful — a stable-looking score hides churn underneath |
| Placements lost to a competitor’s content refresh | Common — citations are contestable, not banked |
| Cadence needed to hold a position | Weekly monitoring; monthly is too slow for fast niches |
A citation you won this month can be gone next month — displaced by a competitor’s refresh, a model update, or a re-crawl. Our tracking puts the median half-life at roughly three to six weeks before a given placement churns. That is our number, from our monitoring stack, and it varies by niche.
The implication is direct — a one-off AEO win is not a stable asset. It is a position you have to defend. This is the whole argument of citation half-life, and it is why measuring AI citations on a weekly cadence is non-negotiable. A monthly report on a fast niche is a report on last month’s reality.
What the numbers mean for budget allocation
Stats are only useful if they change a decision. Here is how this set should change yours.
Adoption is past the tipping point. The 55–70% B2B-adoption range means AEO is no longer an experiment line item — it is a defensive necessity. If your competitor is cited and you are not, the buyer never sees you.
Scarcity raises the price of waiting. Three to eight citation slots, not ten. Every quarter you wait, a competitor banks compounding citations on the highest-volume prompts — the math behind why you should act now.
Split budget by buyer, not by platform volume. ChatGPT’s 55–65% volume share does not mean 60% of your budget goes there. Weight by where your specific buyers research.
Budget for maintenance, not just launch. The 3–6 week half-life means the build is half the cost. Defending the position is the other half. A programme with no maintenance line is a programme that decays.
The cheapest credible first move is a baseline — find out where you stand before you spend. That is what the AI visibility audit delivers.
Methodology note — read this
This matters, so it gets its own section.
This post is not a meta-analysis of named third-party reports. We have not aggregated a stack of branded studies and averaged them. Doing that honestly requires citing each source precisely, and we are not going to attach fake precise figures to real named reports.
Instead, every number here is one of three things:
Publicly stated directional facts. ChatGPT having hundreds of millions of weekly users; Google rolling out AI Overviews broadly across 2024–2025; AI search volume growing fast through 2025. These are general knowledge. We state them as such, without a fake precise citation.
Industry estimates. Broad ranges that reflect the rough consensus across the AEO/GEO field — stated as ranges, never as false-precision single numbers.
Our own modelled ranges. Figures drawn from the 50+ AEO/GEO retainers we have run — pipeline share, citation half-life, AI Overview presence in tracked verticals, platform split. These are labelled “our tracking” or “modelled”. They are honest estimates from real engagement data, not a survey, and not a universal benchmark. Your niche will differ.
So cite this post — but cite it accurately. The value here is the structure and the honest framing, not false precision. If you need numbers specific to your domain, the only honest source is your own baseline. We run that as the AI visibility audit, and the framework behind every number above lives in measuring AI citations.