I've been telling clients the truth about GEO. Here's what I'm saying

The marketing world has lost its mind over Generative Engine Optimisation. Every agency in Australia rebranded last year. Meanwhile, my clients keep asking the same question: do we need to spend more on this stuff?

So I went looking for the data. What I found surprised me, and it's changed how I'm pitching every new project this quarter.

The gap between investment and reality is enormous

Here's the figure that broke my brain. AI tools currently account for less than 2% of total desktop web visits in the US and Europe, per the latest Datos data from Q1 2026. Yet 67% of enterprise SEO budgets have been reallocated to GEO line items, according to GenOptima research. Money is flooding into a channel that barely exists yet.

Google, by the way, just bounced back to 94.3% of US search share. ChatGPT processes around 2.5 billion prompts a day, but Search Engine Land confirmed Google still sends 190 times more website traffic than ChatGPT. So the migration story we keep hearing is coming, but slowly.

Then there's the awkward part nobody at conferences mentions. When Google removed a parameter from its search API in September, Reddit's appearance rate in ChatGPT responses dropped from 60% to 10% within weeks. ChatGPT's Wikipedia citations halved. So a tweak to Google's plumbing reshaped ChatGPT's outputs overnight. Britney Muller put it bluntly on LinkedIn: every URL you see in an LLM output came from a search engine API. The "kill SEO, do GEO" pitch is logically backwards.

Lily Ray published a piece in March titled "Your GEO Strategy Might Be Destroying Your SEO." After spending the last six months reviewing client data, I agree.

What's genuinely new is worth your attention

Now, I'm not saying nothing has changed. Three things have shifted in ways that matter for any business serious about being found.

First, off-page signals matter more than ever before. Ahrefs studied 75,000 brands and found YouTube mentions correlate with AI Overview visibility at 0.737, the strongest signal of anything they tested. Branded web mentions came in at 0.664. Backlinks alone landed at just 0.218. Mike King at iPullRank now calls this Relevance Engineering, and he's right. Where your brand shows up across the open web shapes how AI models talk about you.

Second, structured data has gone from nice-to-have to load-bearing. Semrush ran a controlled test and found GPT-4's extraction accuracy jumped from 16% to 54% when proper schema was present. That's roughly a tripling of pickup rate. Then there's the AI Overview side: pages with valid structured data are 2.3 times more likely to appear. If your developer is still calling schema "optional," you've got a conversation to have with them.

Third, where content sits on the page matters now. Roughly 44% of LLM citations come from the first 30% of an article. Then there's the structural finding from Growth Memo: pages with claim-rich introductions get cited 2.1 times more. The optimal extracted passage length sits between 40 and 75 words. So putting a clear, factual answer right under each heading is no longer just good writing. It's structural optimisation.

What I'm telling clients to do this year

Here's my honest pitch when someone asks if they need a separate GEO retainer. For most Canberra SMEs, no. Not yet.

Instead, here's the work that genuinely moves the needle right now. Get your entity profiles complete. That means Wikipedia where appropriate, Crunchbase, your LinkedIn company page, G2 if you're software, all properly cross-linked through schema. Most of the agencies pitching expensive GEO packages skip this step because it's unsexy.

Then sharpen your earned media. Omniscient Digital's analysis of 23,000 citations found earned media supplies 48% of AI citations against just 23% from owned content. So one Forbes or HBR placement now does double duty as both ranking authority and AI corroboration signal. Expert-quote platforms like Featured.com, which absorbed HARO last year, are back to being serious infrastructure.

After that, audit your structured data and your content structure. Run your top 20 pages and check whether the answer to the page's main question appears in the first 75 words. Then add FAQ schema, Article schema, Organization schema with sameAs properties pointing to your other profiles. Boring work. Genuinely effective.

Finally, be honest about measurement. The biggest research finding of the year, from Rand Fishkin's 2,961-prompt study with SparkToro, is that AI tools return different brand recommendations almost every single run. There's less than a 1 in 100 chance the same prompt produces the same brand list twice. So anyone selling you a "ranking position in AI" dashboard is selling you theatre. Track frequency of mention over time instead. Add a "how did you hear about us" question to your lead forms. That's where ground truth lives now.

The contrarian take

Here's what I'm watching closely this year. Every Australian agency, including ones I respect, has rebranded around GEO since November. None has published proprietary research at the level of the practitioners I've quoted above. We're consumers and repackagers of US thought leadership.

That's both the credibility problem and the opportunity. The agency that runs and publishes a serious Australian GEO study, with disclosed methodology and AU-specific verticals, will own a discourse vacuum no one else has filled.

Until that happens, my advice is simple. Do SEO properly. Do the entity and citation layer that always sat alongside SEO at higher levels of practice. Measure honestly. Don't pay $5,000 a month for a dashboard you can't explain back to your own clients.

The brands winning AI citations in 2026 are the brands that won search in 2024. The work hasn't changed as much as the marketing around it.