An electrician asks an AI assistant a test question: “How much does it cost to install a backup inverter system for a three-bedroom house in Johannesburg?” The answer is detailed, helpful and quotes a competitor’s website twice. It mentions typical price ranges, how long installation takes and what affects the cost, all drawn from the competitor’s guide.
The electrician’s own website has a services page that says: “We offer professional inverter and solar installations. Quality workmanship guaranteed. Contact us for a quote.” It’s accurate. It’s also almost impossible for an AI system to quote, because it contains nothing specific to quote.
As more customers ask AI assistants for advice before they contact a business, a new question has emerged alongside traditional search optimisation: what makes content more likely to be cited in an AI-generated answer? The first serious academic research on the topic offers some practical clues.
One Like insight: AI search quotes specifics. Vague claims give it nothing to work with.
What the research found
In 2024, a team of researchers from Princeton University, IIT Delhi and other institutions, led by Pranjal Aggarwal, presented a paper at KDD, one of the world’s leading data-science conferences. It was titled simply “GEO: Generative Engine Optimization”, and it coined the term now widely used for optimising content to appear in AI-generated answers.
The researchers built a benchmark of around 10,000 real search queries across many subjects. They then tested how different changes to the content of source web pages affected how often and how prominently those pages were cited in AI-generated answers. They found that including citations, quotations from relevant sources, and statistics could significantly boost a source’s visibility, by up to 40% in their tests. When they tried the same techniques on Perplexity, a real AI search engine, visibility improved by up to 37%.
Not every technique helped. Simply stuffing pages with keywords, a tactic associated with old-fashioned search optimisation, performed poorly. Making content clearer and easier to read also helped, though less than adding concrete evidence. And the researchers found that the most effective methods varied by subject area.
The usual caveats apply. These were controlled experiments with specific AI systems at a particular point in time, and AI search changes rapidly. No technique can guarantee your business will be cited. But the direction is consistent with how these systems work: they look for content that answers a question clearly and can be backed up.
Writing content AI can cite
The study’s findings translate into a few practical habits for small business websites.
Answer real questions directly. Create pages or sections that answer the specific questions customers ask: how much something costs, how long it takes, what affects the price, what the options are. Put the answer near the top, in plain language.
Add specific, honest numbers. Typical price ranges, timeframes, sizes, success rates or the number of jobs completed. “Most three-bedroom inverter installations take one day and typically cost between R45,000 and R90,000, depending on battery size” is far more quotable than “contact us for a quote”. Only publish numbers you can stand behind.
Reference credible sources. Where relevant, link to regulations, standards, manufacturer guidance or reputable research. It shows your information is grounded and gives AI systems something to verify against.
Include genuine expert perspective. A short quote or explanation from the person doing the work, such as the qualified electrician or the registered dietitian, adds authority and a human voice.
Write clearly. As we explored in Plain Language Isn’t Dumbing Down, clear writing helps people, and it seems to help AI systems too.
These habits overlap heavily with what makes content useful for human readers, which is reassuring. They are good practice whatever happens to search.
[Infographic placement] Vague versus citable. Two versions of the same service page. Vague: “Professional installations. Quality guaranteed. Contact us for a quote.” Citable: “Most three-bedroom installations take one day and cost R45,000–R90,000, depending on battery size. All work meets SANS 10142 wiring standards. — Thabo M., qualified electrician, 15 years’ experience.” Caption: Aggarwal et al. (2024): citations, quotations and statistics boosted visibility in AI answers by up to 40%.
Where to start
List the ten questions customers most often ask before they buy from you. For each, check whether your website answers it clearly, with specific information rather than an invitation to call. Then rewrite or create the pages that don’t, adding honest numbers, relevant references and a short expert explanation where it helps.
Then test the result. Ask two or three AI assistants the same questions your customers would ask, as we suggested in When Google Answers for You. Note whether your business appears, and which sources do. Repeat the test every few months, because these systems change.
A few questions can help:
- What do customers ask most often before they choose us?
- Do our pages answer those questions with specifics, or only with general claims?
- Where could we add honest numbers, credible references or expert explanation?
- When we ask AI assistants our customers’ questions, which sources do they quote?
Which brings us back to the electrician and his competitor’s twice-quoted guide. The competitor wasn’t necessarily better at electrical work. He had simply written down, clearly and specifically, the things customers want to know. A single, honest guide to inverter installation costs and timeframes, written by someone who does the work every week, could put the electrician’s own expertise where AI search can find it.
Sources
- Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, “GEO: Generative Engine Optimization”, Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’24), 2024.
