Why does AI analyse you — but not recommend anything to you?

The technical aspect GEO plays a major role in determining whether AI references your page — and has nothing to do with whether it recommends your business. Harvard Business Review’s SPRINT principles translated into AI visibility.

Technical GEO makes you readable to AI. It does not make you the chosen one. In our own survey of 465 Danish companies, the technical score was a strong predictor of whether a page was used as a source — but said nothing about whether the company was recommended. The other factor is determined by the very same thing that has always determined a sale: that the customer can see you’re solving their exact problem. Harvard Business Review has summed it up in six letters, and they translate surprisingly directly into AI visibility.

Why are you being read by AI — but not recommended?

Because they are two different outcomes, and most people treat them as one and the same. When we asked four AI engines 576 customer questions about 465 Danish companies, the results were sharply divided: the pages the engines cited as a source had a technical score of 77.9, compared with 72.7 for the others (p=0.000). The companies recommended by the engines scored 74.9 compared with 74.9 — no difference at all (p=0.988). The full analysis can be found in ‘What determines whether AI mentions your company’.

The closest external benchmark points in the same direction. Ahrefs compared 75,000 brands against their visibility in Google AI Overviews (published 26 May 2025) and found that mentions of the brand were more closely correlated with visibility (rank correlation 0.664) than the number of backlinks was (0.218). The analysis can be found here. It’s a single dimension — Google AI Overviews, not ChatGPT or Perplexity — and correlation does not imply causation. But the direction is the same: what lies outside your HTML carries significant weight when it comes to selection.

In short: the technical aspects are your ticket in. They determine whether the model can read your content. It doesn’t determine whether the model points to you when a customer asks, ‘Who should I call?’. For that, you need to be clear on precisely the points on which a buyer will judge you anyway — and research has actually been done on this.

What is SPRINT?

SPRINT is a sales framework from the Harvard Business Review, based on interviews with more than 250 founders worldwide. Dave Rubinstein (founder of 100 Founders, formerly of Salesforce and Outreach) and Vincent Onyemah (professor of sales and marketing at Babson College) published it on 24 June 2026 under the title ‘Startup Founders Need a New Sales Playbook’. Their key finding: most companies confuse curiosity with purchase intent and therefore focus their energy in the wrong place.

The six letters stand for Speed, Problem, Results, Implementation, Niche and Trust. The framework is written about people selling to people. It is not GEO research, and the authors say nothing about AI search. But every principle is about removing the buyer’s uncertainty — and a language model tasked with selecting three companies out of thirty faces exactly the same challenge and exactly the same lack of information.

How do you translate SPRINT into AI visibility?

By asking what each principle means when the reader isn’t a patient human being, but a model that’s skimming your raw HTML and has to make a choice. Below, Rubinstein and Onyemah’s principle is listed first, followed by our translation.

S — Speed: make yourself understood in the very first lines

Can a visitor see within a few seconds that you understand their situation exactly? Not “We’re accountants”, but “You’re busy, and you don’t have time to mess about with the annual accounts.” For an AI crawler, the requirement is even stricter: if the answer isn’t in the raw HTML, it doesn’t exist. The model won’t scroll down or make guesses.

P — Problem: name the problem, not the service

Can you articulate the customer’s problem more clearly than the customer can themselves? Specific problem statements, rooted in a concrete context, drive customers to make a purchase. This is also what makes a page quotable: AI engines extract paragraphs that answer a question directly — and a question is easier to match than a service description.

R — Results: give a figure you can stick to

“You’ll have your annual accounts ready within 10 working days, at a fixed price agreed in advance” beats “I can help with your accounts”. Figures are also what models prefer to quote, because they can be reproduced without interpretation. But only promise what you can deliver and measure: a specific figure you can’t meet costs you more trust than a vague promise — and online, it stays there and gets quoted long after you’ve forgotten it.

I — Implementation: address the objection before it’s raised

The real friction in 2026 is the buyer’s fear: will it be a hassle, expensive, or the wrong choice? Address the risk before the customer mentions it. An FAQ that answers the questions customers actually ask does two things at once — it dispels people’s doubts and provides the model with ready-made question-and-answer pairs to draw on.

N — Niche: one buyer type beats ten

Founders who try to sell to everyone end up reaching no one. Start with one buyer persona, one problem, one sales pitch. This is doubly true for AI responses: the customer’s question is almost always specific (“an accountant for a small craft business”), and a page that matches that specific wording outperforms a generic landing page.

T — Trust: trust lies outside your own site

In founder-driven sales, the founder themselves is the trust mechanism — SME customers buy from people, not brands. This is also where technology stops being able to help you. In our analysis, 4,034 sources were cited, and only six domains appeared across all three sectors: krak.dk, findformig.dk, Trustpilot, cvrexplorer.com, LinkedIn and degulesider.dk. All six are directories and registers — in other words, places where people other than yourself say something about you. 294 of the 316 citations to them came from Perplexity, which is the only one of the four search engines that performs its own searches; the other three ran via a shared search plugin that linked back to the companies’ own websites.

What can’t SPRINT do for you?

It cannot be proven to influence AI recommendations, and this must be stated plainly. SPRINT is sales research, not GEO research: Rubinstein and Onyemah have investigated what convinces a buyer, not what makes a language model mention you. We use this as an analogy because the two tasks are similar — not because anyone has measured it.

Our own figures can only tell half the story. They show with reasonable certainty what does not determine whether you’ll be recommended: the technical score, amongst sites that are already visible in search results. They cannot tell us what does. We haven’t measured reviews, age, company size or marketing budget, and any explanation as to why solicitors perform better than accountants would be a guess on our part.

What you get from SPRINT, therefore, is not a shortcut to the AI answers. It’s a sequence: get the technical aspects right so that you can even be found — and then focus your energy on what makes you worth choosing. How to improve your GEO Score covers the first part, and the difference between SEO and GEO explains why the order matters.

Your SPRINT checklist for next week

Six questions you can answer without anyone’s help. If the answer to any of them is ‘no’, that’s where the work lies — and they’re sorted so that the cheapest is at the top.

• Speed: Does your homepage start by describing the customer’s situation — and is that text available in the HTML without JavaScript?
• Problem: Have you noted down the reason that prompts a customer to search within your industry?
• Results: Is there at least one specific, time-bound figure on your site that you can commit to?
• Implementation: Does your FAQ address the three objections you encounter most frequently?
• Niche: Can you describe your ideal customer in a single sentence?
• Trust: Is your business listed in the directories that search engines actually reference?

Frequently Asked Questions

What is the SPRINT framework?
SPRINT is a sales framework from the Harvard Business Review, published on 24 June 2026 by Dave Rubinstein and Vincent Onyemah, based on interviews with more than 250 founders. The letters stand for Speed, Problem, Results, Implementation, Niche and Trust, and the key finding is that most companies confuse customer curiosity with purchase intent.

Does SPRINT help you get mentioned by AI?
This hasn’t been measured, and we don’t claim that it does. SPRINT is sales research, not GEO research. We use it because a language model tasked with selecting three companies out of thirty faces the same lack of information as a buyer — but no one has investigated whether the six principles influence AI recommendations.

Why is technical GEO not enough to warrant a recommendation?
Because technical merit and recommendation are two different outcomes. Across 465 Danish company websites (Geoa, 26 July 2026), the cited pages scored 77.9 versus 72.7 (p=0.000), whilst the recommended companies scored 74.9 versus 74.9 (p=0.988). The ‘technique’ therefore strongly determines whether your page is used as a source — and says nothing measurable about whether your company is actually referenced.

What should I do first as an SME?
Finalise the technical aspects so that your content can be read at all: the core content in the raw HTML, accessibility for AI crawlers in robots.txt, and structured data. Then focus your efforts on the six SPRINT points. The order matters — most people start on the second point before the first is in place, and so their content is never read.

Where do the figures in the article come from?
The p-values and the 4,034 cited sources are Geoa’s own survey from 26 July 2026 (465 companies, 576 AI responses) — the methodology and caveats are set out in the full analysis, including the fact that only Perplexity of the four search engines conducted the search itself. The 75,000 brands are from Ahrefs’ correlation analysis dated 26 May 2025 and cover Google AI Overviews alone. The 250+ founders are from the Harvard Business Review, 24 June 2026.