The AI Search Grift, One Year Later: They Sold Each Other the Territory

Published July 2026.

A year ago we wrote that the "Generative Engine Optimization" industry was mostly repackaged SEO sold with manufactured urgency. The 2026 update is stranger: the market didn't punish the pitch — it priced it. In 2025 they sold you the map. In 2026 they sold each other the territory.

One housekeeping note first, because this piece is about evidence standards. Our 2025 article opened with the story of a Toronto e-commerce company that paid $50,000 to a GEO consultant and got nothing. That anecdote circulated widely — and it traces only to vendor blogs, with no named victim, contract, or filing. We can't verify it, so we're retiring it. That's the point, though: this entire category runs on unfalsifiable stories, in both directions. The sellers have case studies nobody can check; the skeptics have horror stories nobody can check. What follows is only what can be checked.

The money got real

Whatever you think of the evidence, the capital has voted.

On April 28, 2026, Adobe completed its acquisition of Semrush (Adobe newsroom, Apr 28, 2026) — roughly $1.9 billion at $12.00 per share, per the November 2025 announcement terms — and positioned the deal explicitly around brand visibility in AI surfaces: SEO, GEO, and "agentic search optimization" as one product line. Adobe's own analytics, cited in the same release, claim AI traffic to US retail sites rose 269% year over year as of March 2026. That's Adobe's number, from Adobe's panel, marketing Adobe's acquisition — but the deal itself is a fact.

In February, Profound — the best-known AI-visibility tracker — raised a $96 million Series C at a $1 billion valuation led by Lightspeed, with Sequoia and Kleiner Perkins participating, taking total funding past $155 million (Fortune, Feb 24, 2026). The company claims 700+ enterprise customers including more than 10% of the Fortune 500; those counts are Profound's own. Berlin's Peec AI announced $10 million ARR sixteen months after launch (company release, May 28, 2026), opened its first US office in May, and is reportedly raising at around a $200 million pre-money valuation (Sifted, citing sources on an unclosed round).

And Microsoft quietly made the strongest anti-pricing argument in the category: in spring 2026 it expanded Microsoft Clarity — free — to show site owners which of their pages get referenced in AI-driven answers and how often (the Citations dashboard reached general availability on May 13, 2026, after an earlier preview). When a hyperscaler gives away the measurement layer, the $399-a-month "share of model" dashboard has a positioning problem.

So the category your 2025 skeptic called scam-adjacent now has a unicorn, a $1.9 billion big-tech exit, and free big-tech competition. The money got real. The question is whether the evidence did.

The denominator still rounds to zero

Here is the traffic picture the pitch decks skip.

Similarweb-derived 2026 estimates put AI platforms at roughly 0.15–0.3% of global internet traffic, against roughly 43–48% from organic search; AI chatbot referrals remain under 1% of total publisher traffic even after growing several-fold from a near-zero base. Two disclosures travel with that stat: it's panel-based vendor data, and Similarweb itself now sells GEO tooling — the sharpest anti-hype numbers in the category are published by a company monetizing the hype. Read them as estimates from an interested party that happen to cut against its own product line.

Meanwhile the search everyone actually uses keeps sending fewer clicks anywhere. Rand Fishkin's June 8, 2026 study of Similarweb clickstream data estimates that 68.01% of US Google searches ended without a click to the open web in early 2026, up from 60.45% in 2024 (SparkToro/Similarweb; Fishkin himself cautions the multi-year comparison mixes panels). And the AI surfaces are genuinely enormous as surfaces: Google said at I/O 2026 that AI Mode passed 1 billion monthly users with query volume more than doubling each quarter (Google, May 2026).

Hold those together and you get the honest 2026 picture: AI answers are where attention increasingly sits, and they are terrible at exporting that attention to your website. Both things are true. The grift lives in the gap — selling billion-user reach while your analytics show a fraction of a percent of visits.

What the evidence actually supports

In May 2026 the field finally got something like a systematic review. Cyrus Shepard (Zyppy, May 7, 2026) synthesized 54 experiments, patents, and case studies into 23 scored "AI citation factors." The top of the table: URL accessibility (9.5), traditional search rank (9.4), fan-out rank (9.3), preview control (9.2), query-answer match (9.2). The bottom: llms.txt — a staple of GEO vendor pitches — scored 2 out of 10, with Shepard "unable to find any credible evidence" it influences citations. His own disclaimers apply: these are not confirmed ranking factors, correlation is not causation, and the scoring is one researcher's judgment. But note what the best available synthesis says: the strongest predictor of getting cited by AI systems is ranking well in ordinary search. The sold-separately discipline's top factor is the old discipline.

Then there's the measurement problem. Per Profound's own published research, 40–60% of the domains cited by major AI platforms change month to month. Sit with that: at churn that high, any before/after case study is unfalsifiable. Your citations went up after the engagement? They might have anyway; half the citation set turns over monthly. They went down? Same answer. A market where the vendor's own data makes the vendor's case studies untestable is a market running on faith.

The tools don't agree with each other either. Digiday's May 1, 2026 reporting had agency executives on record that different AI-visibility trackers return different answers to identical prompts, that pricing runs from $99 to $399 a month for the leading tools (with service layers reportedly up to $1,000), and — from one agency CEO — that "we don't know enough of these companies to be charging what they're charging." When the trade press frames the question as "a strategic necessity or the latest ad tech grift," the vibe has shifted.

The platforms built the anti-fraud plumbing

The most telling 2026 development isn't a study. It's infrastructure.

On May 15, 2026, Google extended its Search spam policies to explicitly cover "attempting to manipulate generative AI responses in Google Search" — AI Overviews and AI Mode — with penalties from demotion to removal (Google Search Central; Search Engine Land, May 15, 2026). A meaningful slice of what's sold as GEO tactics is now, formally, spam.

In early June 2026, Google updated its long-standing "Do you need an SEO?" documentation to do something it had never done: tell business owners to file FTC complaints about fraudulent SEO services. The same update added AEO/GEO as a recognized service category — while warning about spam risk inside it — and cautioned against over-relying on third-party SEO-tool data (Google Search Central, "Do you need an SEO?", updated early June 2026).

Be precise about what this is and isn't. As of this writing we could find no named FTC action, lawsuit, or court record against a GEO vendor. The 2026 story is plumbing, not prosecutions: complaint routing, policy definitions, enforcement hooks. Whether anything flows through the pipes is a checkable prediction for the 2027 follow-up.

And note the quiet irony on the paid side. The only officially sanctioned way to buy your way into an AI answer surface in 2026 is an ad — labeled, disclosed, and (per the platform) walled off from the answer itself. OpenAI began testing ads in ChatGPT for US free tiers on February 9, 2026, reached the UK in the first week of June, and its stated principle is blunt: "Ads do not influence the answers ChatGPT gives you" (OpenAI). Nobody can sell you organic placement in an answer; no platform has endorsed any third-party optimization service. (How widespread those ads are is itself unmeasurable from outside: independent trackers put ad-carrying US responses anywhere from roughly 20% to 49% in May 2026 depending on methodology and week — llmrefs, Writesonic and cloro.dev all measured different numbers. No authoritative figure exists.)

What the doom narrative gets wrong

Fairness pass, because the pure "it's all a scam" take doesn't survive the data either.

People who use AI answers do click — more than the discourse says. The Reuters Institute's Digital News Report 2026 found weekly AI chatbot use for news rose from 7% to 10% globally, and — the stat that cuts against the doom narrative — 42% of those chatbot news users say they always or often click through to the original source, ranging from 56% in South Korea to 26% in Denmark (Reuters Institute, June 2026). That's self-reported behavior from a survey, not clickstream, and self-reports flatter. But it complicates the "nobody ever clicks" story: among people who have actually adopted the tools, the claimed click-through habit is substantial. Small base, real engagement.

Some of what GEO vendors recommend is just good practice. Zyppy's top factors — accessible URLs, ranking content, controlling your snippets, matching queries to answers — are things worth doing whether or not AI search exists. If a consultant's "GEO audit" makes you fix crawlability and write pages that actually answer questions, you got real SEO at a markup.

Measurement itself is now free. Microsoft Clarity's AI-visibility reporting costs nothing. Watching whether AI referrals matter for your business is cheap and worth doing; that's not hype, it's instrumentation.

The scam was never "AI search is fake." It's real and growing. The scam is the pricing of certainty nobody has — guarantees against 50% monthly churn, proprietary methodologies whose best-evidenced ingredient is "rank in Google," and urgency invoices against a channel that is under 1% of your traffic.

What to actually do (advertiser edition)

We run an ad-intelligence product, so here is the practitioner version, no acronyms required:

1. Measure before you spend. Segment AI referrals (ChatGPT, Perplexity, Gemini, Copilot) in your analytics and turn on Microsoft Clarity's free AI-visibility reporting. For most advertisers this will show a sub-1% channel. That number — not a vendor's "share of model" score — is your budget ceiling for the category.
2. Refuse guarantees, demand baselines. With 40–60% of AI citations churning monthly (Profound's own data), any vendor promising specific citation outcomes is selling weather forecasts by the season. If you engage anyone, require a pre-engagement citation baseline, a holdout of comparable pages, and month-over-month tracking — or don't engage.
3. Spend the "GEO budget" on ranking. The best-evidenced predictor of AI citations is traditional search rank (Zyppy, May 2026). Money aimed at AI visibility is better spent on the content and technical work that ranks — which you can measure with tools you already have.
4. Don't buy tactics Google now defines as spam. If a pitch involves manipulating AI answers — recommendation-poisoning listicles, prompt-injection tricks, mass "AI-optimized" pages — that's inside Google's May 15 spam policy, and Google's own guidance now points defrauded buyers to the FTC. The downside is no longer just wasted budget.
5. If you want to be in AI answers with certainty, that's called an ad. ChatGPT ads exist in a growing list of markets; Google's AI Mode carries Google's ad formats. Labeled placement with reported spend beats unverifiable "organic influence" — and unlike GEO invoices, ad spend shows up in platforms where you can track changes, anomalies, and what actually drove results.
6. Re-check quarterly, not never. The Reuters 42% click-through figure and 10% adoption say this channel's users are engaged and multiplying. Sub-1% channels sometimes become 5% channels. The correct posture is cheap instrumentation now, real budget when your own data says so.

Last year the pitch was that you were about to be left behind. This year the people making that pitch sold their companies, raised their unicorn rounds, and priced their dashboards — while the referral traffic they were selling protection for still rounds to zero, and the best independent evidence says the way into AI answers is the boring work of ranking. The map got acquired for $1.9 billion. The territory is still under 1% of your visits.

Save the retainer. Instrument the channel, do the boring work, and read the vendor's own churn data back to them when they call.

Works Cited

• Adobe. "Adobe Completes Semrush Acquisition." Adobe newsroom, 28 Apr. 2026. (Deal terms from the Nov. 2025 announcement.)
• Fortune. "As AI Threatens Search, Profound Raises $96 Million to Help Brands Stay Visible." 24 Feb. 2026.
• Sifted. "Berlin's Peec AI Targeting $200m Valuation in New Fundraise, Sources Say." 2026; Peec ARR: company release via GlobeNewswire, 28 May 2026.
• Fishkin, Rand. "In 2026, Less than One Third of Google Searches Still Send a Click." SparkToro, 8 June 2026.
• Shepard, Cyrus. "AI Citation Ranking Factors." Zyppy Signal, 7 May 2026.
• McCoy, Kimeko. "Marketers Question Expensive AI Visibility Tools as Inconsistent Results Fuel Skepticism." Digiday, 1 May 2026.
• Google Search Central. Spam policies (updated 15 May 2026); "Do you need an SEO?" (updated early June 2026).
• OpenAI. "Our Approach to Advertising and Expanding Access"; "Testing Ads in ChatGPT." Jan.–Feb. 2026.
• Reuters Institute for the Study of Journalism. Digital News Report 2026 (executive summary). June 2026.
• Similarweb. Gen-AI traffic statistics, 2026 (vendor panel data; Similarweb also sells GEO tooling).
• Profound. Citation-volatility research (40–60% of cited domains change month to month). 2026.
• Google. I/O 2026 Search announcements. May 2026.
• Microsoft. Microsoft Clarity Citations dashboard (AI visibility); generally available 13 May 2026.