George Chasiotis owns Minuttia, an SEO and content agency for B2B software, and spends his time on strategy and research rather than day-to-day delivery. That research is why I wanted this conversation: his team runs data studies on AI search, so the claims come with numbers attached.
The headline finding is a strange one. They set out to test whether a SaaS company’s G2 rating correlated with its visibility in AI search. It didn’t — they couldn’t prove it. What they could prove was a high correlation with the number of reviews.
First, the myth worth killing
The idea in circulation is that SEO is finished and AI search is the thing to work on now. Minuttia’s August study points the other way: there’s a very high correlation between ranking in Google’s top ten and appearing in AI search — AI Overviews, ChatGPT, Claude.
The two overlap heavily. George’s read on why that isn’t widely understood is uncharitable but hard to dismiss: a wave of agencies has moved into AI search opportunistically, without SEO fundamentals, under pressure to show results fast. The tactics that follow often work — which is the problem.
“It works until it doesn’t.”
And some of them carry brand consequences that outlast the ranking they bought.
What he proposes instead is a three-layer model.
Layer 1 — brand experience
The one people like least, because they control it least.
Brand experience is the accumulation of small moments: how the product behaves, what support is like on the day someone needs it, how you communicate a pricing change. His argument is simply that the truth gets out, and what gets out forms a perception of who you are and whether you deliver what you claim.
Which brings us back to that G2 finding. Review count correlates with AI search visibility; rating doesn’t. A company can sit at 832 reviews and a 3.2 average with people complaining about pricing — the count isn’t measuring happiness. It’s measuring how much impact the brand has in its marketplace.
His worked example is Meilisearch, a client: search-engine-as-software, open source, tens of thousands of GitHub stars, hundreds of documentation pages, and a community of genuine fans. Doing SEO and AI search work for them isn’t easy, but it’s easier, because the models already have enormous context about what the product is and who it serves.
The forecast underneath this is the part worth arguing about:
Google rewarded mediocre content and average brand experiences. AI search won’t.
If that holds, “me too” B2B SaaS has a harder road than it did, because the channel that used to let you compensate with volume no longer does.
Layer 2 — owned perception
Everything you have editorial control over: your website first, then social handles, YouTube, Crunchbase, anywhere you can publish and edit at will.
The useful finding here is what their study calls evaluation attributes. When AI search engines answer a commercial prompt by recommending software, they reason over a consistent set of attributes — the obvious ones like features, integrations, pricing and ease of use, but also more specific ones: team collaboration, AI capabilities, scalability.
The move follows directly. Identify the evaluation attributes for your category, then publish a page for each one. You are supplying the ground truth the model reasons over.
And this cuts both ways, which is the sharpest story in the episode. Minuttia published a page explaining that they don’t send blind proposals without an intro call. From their own testing, that page alone loses them leads — because when someone prompts for agencies that will return ten to twelve proposals by Friday, the model knows Minuttia won’t, and leaves them out.
Worth sitting with. Your own pages don’t just help you get recommended; they also tell the model when to rule you out. A page written for one audience can quietly disqualify you with another.
Layer 3 — multi-channel perception
The surfaces you don’t control: Reddit threads, review sites, listicles, other people’s videos. Sentiment at the extremes — strongly positive or strongly negative — tends to carry more weight.
It’s multi-format as much as multi-channel. Their studies show YouTube is cited far more in Google’s products (AI Overviews, Gemini) than in ChatGPT, which is unsurprising once you remember who owns YouTube.
On the listicle question everyone asks, their unpublished study on self-promotional content found that for alternatives prompts and best X prompts, third-party and self-promotional listicles are prominently used as citation sources across most engines, ChatGPT excepted. He’s careful to flag that his agency does this work, so he’s not a neutral party, and that working today isn’t the same as working in two years.
His brand-safety test is the practical takeaway. He described an agency that set up a YouTube channel publishing AI-generated listicle videos that conveniently ranked their client first. It works — until a competitor finds it and writes the Reddit thread about it.
Two years from now, would we stand behind this?
On backlinks he lands somewhere sensible: probably not very relevant to AI search directly, but if they still help you rank in blue links, and blue-link visibility correlates strongly with AI visibility, then they matter transitively. Less than five years ago. Not zero.
You can’t be HubSpot
I asked the question that nags at anyone in a crowded category: HubSpot shows up ten times out of ten when you ask for a CRM. Where does that fit?
His answer was blunt — being HubSpot is taken. The premise of competing head-on is already wrong.
What’s available is specialisation: the HubSpot of agencies, the best CRM for a specific region or a specific use case. Mature categories have a leader and a long tail, and the long tail is where the opening is. He adds a caveat worth remembering — AI search, like Google, is fundamentally a demand capture channel, so a genuinely new category may have no demand to capture yet.
On AI-written content
He sorts the industry into three camps: the deniers (no AI content, ever), the absolutists (why employ writers at all), and a middle position.
His view is that the absolutist camp has already lost for anything aimed at a search audience — and that one of Minuttia’s better sources of new business is brands coming out of the honeymoon period, having published four hundred pieces in four months and damaged their site’s prospects in the process.
Minuttia sits in the middle: AI for research and briefing, to make the work better rather than cheaper, with humans writing. As he points out, an agency could quietly automate its own blog and nobody would check. They still pay people to write it.
I put the obvious question to him, since this article is exactly the thing in question: is it fine to turn a podcast into a written piece with AI assistance, when the source is an original conversation?
His answer was that it depends on the objective. For repurposing and distribution, he sees nothing wrong with it. But working with a public company whose content passes through legal review, you can’t be casual about how the output was produced. The test he applies is the same one as above — in two years, would you still say you’d do it the same way?
FAQ
Is SEO dead now that AI search exists? No. Their study found a high correlation between Google top-ten visibility and AI search visibility.
Do G2 reviews affect AI search visibility? The number of reviews correlates; the rating didn’t show a provable correlation.
Do listicles still get cited? Yes, for alternatives and best-of prompts across most engines, with ChatGPT the exception.
How do you compete with the category leader? By specialising — a vertical, a region, a use case — rather than competing head-on.
The takeaway
The uncomfortable implication of all three layers is that most of AI search visibility isn’t an SEO task. Review count, what support is like, whether people have reason to talk about you — those sit outside the remit of the person usually asked to fix visibility.
What is directly actionable is layer two. Work out the evaluation attributes for your category, check which ones you’ve never addressed on your own site, and write those pages. Then read your existing pages once more for anything that quietly disqualifies you.
That’s the kind of thing I’d look at first in an organic program. If you want a straight answer about yours, book a call — and for the full conversation with George, watch the episode or connect with him on LinkedIn.