Is AEO real, or is it the new snake oil?

Part 1, The case against, put properly

Google has said, officially and in writing, that you don't need this

On 15 May 2026, Google published guidance on Search Central covering optimisation for its generative AI features. The substance:

• Google's AI features are rooted in its core Search ranking and quality systems.
• SEO best practice remains relevant and foundational to performing well in those features.
• The terms "AEO" and "GEO" are not necessary.

This was not a leak or an off-hand remark at a conference. It is documentation. Google's Danny Sullivan has separately said that good SEO is good GEO, and that SEO for AI is still SEO. A Google VP has described substantial overlap between the two.

That is about as clear as a platform ever gets.

Google's search advocate has linked the acronym push directly to scams

John Mueller, on Bluesky: the higher the urgency and the stronger the push of new acronyms, the more likely it is that someone is simply making spam and scamming.

Read that as a buyer, not as a marketer. Someone whose job is to explain how Google Search works looked at the AEO/GEO/AIO/VEO wave and concluded that the intensity of the sales pressure was itself the tell.

He was describing a real pattern. I've watched it up close for two years.

The money is being wasted, in documented amounts

The case study that circulates in the sceptic community: a mid-sized Toronto e-commerce business paid $50,000 to a self-described generative engine optimisation expert on a promise to dominate AI search. Six months later, no AI referral traffic, no ChatGPT citations, no AI Overview appearances.

I have no way to verify that particular story. I can tell you it is entirely plausible, because I have seen the proposals that produce that outcome.

Several of the most-sold tactics have no evidence behind them

Cyrus Shepard of Zyppy published a synthesis in May 2026, 23 AI citation factors scored across 54 studies, patents, experiments and case studies, weighted by repeatability, strength of evidence and official platform support. It is the most rigorous consolidation the category has.

llms.txt scored 2.0 out of 10. No credible evidence, in any of the 54 sources, that maintaining one influences AI citations measurably. Google has said separately it won't use it.

It is still being sold. Sometimes as a headline deliverable.

Structured data scored 5.6, and Shepard flags it as contested. Nearly every study that examined the relationship found a positive correlation, but LLMs don't ingest schema as training data, so the mechanism is genuinely unclear. It's probably worth doing. Anyone telling you they know why it works is ahead of the evidence.

The single biggest finding is that this is mostly just SEO

Shepard's own summary of his 23 factors: you don't need an entirely new playbook. The overlap between traditional SEO signals and AI citation signals is substantial. Relevance, trust, topical authority, extractability, all of which align with what good SEO has always meant.

Ahrefs found 38% of AI Overview citations come from Google's top ten, with the overlap strengthening beyond position ten. seoClarity found 97% of AI Overviews cite at least one top-20 result.

Win organic search and you are most of the way to winning AI citations. That is the finding. It does not support a separate discipline with separate pricing.

Scaling AI content is an actively dangerous tactic being sold as the solution

This one deserves particular attention, because "unlimited content production" has become a headline feature across the tooling market.

The pattern in public traffic data is consistent and grim. A site goes all-in on AI-generated content. Rankings and traffic climb steeply. Google's spam systems catch up. The decline that follows is longer and deeper than the climb, and recovery is slow where it happens at all. This has been documented for brands whose leadership publicly endorsed the content-scaling platforms they were using.

Rapid AI content scaling can violate Google's spam policies. A tool selling you volume is selling you the mechanism of your own penalty.

And the channel is still small

AI referral traffic sits at roughly 1% of total website traffic, growing about a point a month. Google's search revenue rose 19% year-on-year in Q1 2026 with query volume at an all-time high. Google holds around 90% of traditional search.

If someone has told you that search as you know it has ended, they have overstated the position by a distance.

Part 3, What the sceptics get wrong

Four things. Each one is load-bearing.

1. Google only speaks for Google

This is the largest gap in the sceptical case, and it is rarely addressed.

Google's guidance is accurate about Google's own surfaces, AI Overviews and AI Mode run on core Search systems, so Google's ranking work carries over. Nobody disputes that.

But ChatGPT doesn't use Google's index. Seer Interactive analysed over 500 citations and found 87% of SearchGPT citations matched Bing's top ten results, against only 56% correlation with Google's. Perplexity runs a proprietary index of over 200 billion URLs and weights community sources, Reddit in particular, far more heavily than Google does. Claude retrieves differently again, and is the fastest-growing referrer in B2B: Goodie tracked 16 brands with sufficient volume and found 14 of them grew their Claude share, with a typical gain of 7.14 percentage points and a probability of that pattern arising by chance of about 0.2%.

The number that settles it: a study across 55,936 queries and six LLM search engines found roughly 37% of the domains those engines cite never appear in traditional search results at all.

Google's advice is correct and covers perhaps half the surface area. Taking Google's word on ChatGPT's retrieval behaviour is like taking Coca-Cola's word on Pepsi's supply chain. Not dishonest. Just not their information to give.

Note also the incentive. Google is a platform with enormous exposure to the possibility that discovery moves off its properties. Its framing will be conservative, and shaped by its own interests, exactly as any platform's would be. That doesn't make the guidance wrong. It does mean you shouldn't treat it as neutral.

2. "Rankings carry over" is true on average and unreliable for you specifically

The sceptical case leans on the overlap figures. Those figures are unstable.

• Ahrefs, March 2026, overlap between Google's top ten and AI Overview citations fell from 76% in July 2025 to 38% in eight months.
• BrightEdge, over a comparable period: overlap rising, from 32.3% to 54.5%.
• seoClarity: 97% of AI Overviews cite at least one top-20 result.
• The Digital Bloom, on AI Overviews specifically: 93.67% of citations link to at least one top-ten organic result.

These are not reconcilable. They measure different query sets, different verticals, different time windows, with different definitions of a citation.

What that means practically: "your rankings will carry over" is a claim with a wide confidence interval and no per-business predictive value. And it says nothing at all about ChatGPT, Perplexity or Claude, which don't use Google's index.

The only way to know your position is to measure your position.

3. The signal that drives recommendation is one SEO retainers were never built to deliver

Here is the finding that changes the shape of the argument.

Seer Interactive ran six behavioural tests across 362,388 AI responses in 2026. Their hypothesis: the model selects which brands to recommend from its trained knowledge first, then retrieves sources to support a decision it has already made. The citation is the bibliography, not the brainstorm.

The corroborating data: Ahrefs studied 75,000 brands and found branded web mentions correlate 0.664 with AI Overview visibility, while backlinks correlate 0.218, roughly a third as predictive. The three strongest signals were all off-site brand signals rather than link metrics.

If that's right, then two things follow that neither the sceptics nor the tooling vendors are saying:

Against the vendors: page-level optimisation cannot rescue a brand the model has never learned. Most of what is being sold as AEO is optimisation of the bibliography. It will not change the brainstorm.

Against the sceptics: "just do good SEO" doesn't cover this either. Traditional SEO retainers optimise pages and acquire links. The signal that predicts AI recommendation three times more strongly is brand presence, entity clarity, genuine notability, presence in the sources these systems read, review consistency, category authority. That is real work with real cost, and it sits outside the scope of almost every SEO contract ever signed.

So the sceptics are right that this isn't a new discipline, and wrong that your existing arrangement already covers it.

4. "It's still just SEO" is true and useless as a purchasing decision

Suppose we accept the strong version: this is entirely SEO, no new discipline, no new playbook.

You still need to know whether AI systems currently recommend you. You still need a measurement method with a stated margin of error. You still need to know which of five platforms you're absent from and why. You still need attribution that separates AI referrals from organic. You still need someone to decide what to do first.

"Do good SEO" is correct advice and not an operating plan. Google is documenting how its systems work. It is not running your marketing.

Calling the work AEO is a labelling convention. I'd happily call it search. The label is not the thing being bought.

How to interrogate any AEO proposal

Seven questions. They're free, and they'll disqualify most of the market.

1. What's the margin of error on your measurement? A single check on a prompt carries ±9–13 points. If there's no confidence interval, you're being shown a screenshot dressed as a trend.
2. Which platforms, and how are you measuring each one? ChatGPT, AI Overviews, AI Mode, Gemini, Claude, Perplexity and Copilot retrieve differently. Coverage of one is not coverage.
3. Show me an intervention that didn't work. Anyone with real data has failures. Anyone with no failures has no data.
4. How much content per month, and what's the originality standard? If the answer is a big number with no gate, that's the penalty pattern.
5. Is llms.txt on the deliverables list? If it's presented as a driver of visibility, they haven't read the evidence.
6. What's your position on Google's May 2026 guidance? If they haven't heard of it, they aren't following the field. If they dismiss it, they're not being straight with you.
7. What are you doing about brand signal, not just page signal? If there's no answer, they're optimising the bibliography.