What Your AEO Agency Isn't Telling You About AI Search
AI search is the biggest organic opportunity in a decade and the hardest channel to measure. Here is what most AEO agencies gloss over when they pitch you.
Samuel Edwards
Every AEO agency pitch deck in 2026 runs on the same two slides. One shows a hockey-stick chart of AI search adoption. The next shows a Perplexity screenshot where the client's brand appears in a cited answer. Then comes the retainer.
What those decks skip is harder to screenshot. AI answer engines do not expose a Search Console. They do not rank deterministically. They send almost no traffic by Google's standards, and when they do, roughly two-thirds of it lands in your analytics as "Direct." A CMO who signs an AEO retainer without pressing on measurement is buying a service whose outputs they cannot audit.
This is not an argument against generative engine optimization. The channel is real, the intent is high, and the brands that establish early citations will compound an advantage. It is an argument for walking in with the uncomfortable questions already loaded.
The Volume Problem Nobody Puts on the Deck
Start with the number that reframes everything. According to Cloudflare Radar data for May 2026, Google sends 87.63% of all search referral traffic, while every AI search engine combined (ChatGPT, Gemini, Claude, and Perplexity) accounts for 0.29% of search referrals. Not 29%. Zero point two nine.
Agencies counter with the growth curve. That curve is real. Perplexity reached roughly 45 million monthly active users by late 2025, up from about 10 million in early 2024, with 780 million monthly search queries and an 85% retention rate. Gartner predicted in February 2024 that traditional search volume would drop 25% by 2026 as users migrate to AI chatbots and virtual agents. Both things can be true: the trajectory matters, and the current base is tiny.
What this means in a buying conversation: an agency promising "AI search traffic" as a near-term revenue line is selling against today's math. Ask them to model expected sessions, not citations. If the number is honest, it will look small. That is not a reason to walk away. It is a reason to price the engagement against long-horizon positioning rather than quarterly pipeline.
Attribution Breaks Before the Click Even Lands
The second thing most AEO pitches gloss over is that a huge share of the traffic they do send is invisible by default. When a user clicks a link inside ChatGPT, Perplexity, or an in-app browser, the referrer header is frequently stripped. GA4 sees no source and files the session under "Direct," the same bucket as someone typing your URL from memory.
The order of magnitude here is not a rounding issue. One attribution analysis found that 70% of AI-influenced visits arrive without a referrer header and get classified as Direct in GA4, with last-click attribution capturing only about 2% of AI search's true revenue contribution. A separate practitioner guide reports 35 to 70% of AI referral sessions arriving without referrer headers depending on platform and device. Google AI Overviews are their own category of invisible: the clicks come from google.com and get filed as regular organic search unless you have custom text-fragment detection in place.
Three questions to ask any AEO vendor before signing:
- Show me the regex you deploy in GA4 to catch Perplexity, Claude, Copilot, DeepSeek, and Grok referrers, and the custom channel group you build around it.
- How do you reconcile branded-search lift and Direct-traffic anomalies against your AI visibility reports? If the answer is "we don't," they are reporting on citations, not revenue.
- What is the self-reported attribution instrument on our site (post-purchase surveys, form fields, sales-call tagging), and who owns keeping it clean?
If the vendor cannot answer those three, you are paying for a monitoring dashboard, not an attribution system. For a broader primer on what separates these disciplines from classic SEO, our breakdown of AEO and GEO vs traditional SEO is a useful second read.
There Is No Search Console for ChatGPT
Google Search Console has conditioned a generation of marketers to expect a query report, an impressions count, and an average position. None of that exists for ChatGPT, Claude, or Perplexity. Position itself is non-deterministic: the same prompt from two users at the same moment can produce different cited sources, different brand mentions, and different ordering.
That volatility is structural, not a bug the agency will fix next quarter. A Semrush topic-authority study of 50,000 brands tracked monthly in ChatGPT from January through June 2026 across 1,094 US categories found that only 15% of AI search categories have a clear brand winner. In the other 85%, visibility is split across many sources and rotates between prompts. If your vendor is showing you a single "AI share of voice" line going up and to the right, ask how many prompts it samples, how often, from what geographies, logged in or out, and across which model versions. The honest answer is that the number is a weighted average over a moving target.
Compounding the problem: being cited is not the same as being named. A Semrush study with Kevin Indig analyzing 3,981 domain appearances found that 62% of AI citations don't lead to brand mentions. ChatGPT cites sources 87% of the time but names the brand in only 20.7% of answers; Gemini names brands 83.7% of the time but cites the source only 21.4% of the time. The implication is uncomfortable for anyone selling "AI citations" as a KPI: a link in the footnote drawer does not necessarily translate into a brand the user remembers, and a brand mention in the prose does not necessarily translate into a click.
The Playbook the Agency Sold You May Not Work
A lot of early AEO guidance treated schema markup as the lever: ship JSON-LD, get cited. The empirical case for that is weaker than most pitch decks admit. An Ahrefs controlled study tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched them against 4,000 control pages, and found adding schema produced no major uplift in citations across Google AI Overviews, AI Mode, or ChatGPT. Schema is still worth shipping for other reasons. It is not an AEO cheat code.
The harder reality is that AI visibility is a topic-level game, won by brands that already have category authority and that produce the kind of structured, comparative, specific content these models prefer to quote. Semrush's 2026 AI Visibility Index analyzed 126 million U.S. AI search prompts across ChatGPT, Gemini, Google AI Mode and Google AI Overviews, and found that only 36 of more than 1,200 brands tracked appeared in the top 100 most-mentioned list on every platform in every month of the study. Cross-platform dominance is rare, concentrated, and bought with topical authority accumulated over years, not sprint cycles.
If your agency's deliverable is "add FAQ schema, write a definition paragraph at the top of each page, resubmit the sitemap," you are buying 2024 tactics. The current frontier looks more like entity consolidation across your content footprint, measurable citation share in a defined prompt set, and the sort of underlying technical hygiene (clean canonicals, no broken 404 pages in cited URLs, consistent entity references) that keeps a model confident enough to quote you.
What a Real AEO Engagement Looks Like
None of this is a case for ignoring AI search. The channel's conversion economics, where you can measure them, are genuinely interesting: multiple B2B datasets report AI-referred visitors converting several multiples above non-branded organic. The point is that the retainer has to be structured around what the channel actually is right now: small-volume, high-intent, poorly instrumented, slow to compound, and governed by topical authority rather than tactical hacks.
A defensible scope of work looks something like this:
- A prompt set of 100-300 category-defining queries, sampled weekly across at least ChatGPT, Perplexity, Gemini, and Google AI Mode, with both citation rate and brand-mention rate reported separately.
- A dual-attribution stack: custom GA4 channel group for AI referrers, plus a self-reported "how did you hear about us" instrument on every conversion path.
- Branded-search and Direct-traffic baselines tracked as proxies, so lift from AI visibility is detectable even when the referrer is stripped.
- Content work prioritized by topic authority gaps, not by schema checklists. See our note on getting your company recommended by ChatGPT for the content shape that actually earns citations.
- A deliberate hedge: traditional SEO work continues at full weight because Google still sends the overwhelming majority of commercial traffic. The AI SEO and GEO program runs alongside, not instead.
Before you sign anything, send over our checklist of questions to ask an SEO company, adapted for AI search: What is your measurement stack? What is your citation-to-click conversion assumption? What happens to our reporting when OpenAI changes how ChatGPT cites sources next quarter? If the answers are vague, the retainer will be too.
Buy the Positioning, Not the Promise
AEO is where SEO was in 2004: a real channel, a thin measurement layer, and a lot of vendors selling confidence they have not yet earned. The brands that win here will treat it like a long-dated option, not a performance channel. They will instrument honestly, accept that reporting is directional, and spend on topical authority that pays off whether the next model release favors citations, mentions, or something nobody has named yet. The agency that tells you otherwise is selling a dashboard. The engagement worth paying for is the one that tells you what it cannot see, and builds anyway.