What the research says about getting cited in AI answers

The study

“GEO: Generative Engine Optimization” by Pranjal Aggarwal and five co-authors was accepted to KDD. The authors built a benchmark of queries across many domains and tested nine ways of rewriting a source page, to see how often an AI search engine’s answer cited and used that page. They also repeated part of the test on Perplexity.

What helped, and by how much

Three changes did best: adding statistics, adding quotations from credible people, and adding citations to reliable sources. The paper reports relative gains of 30 to 40 percent on its position-adjusted word count measure and 15 to 30 percent on a subjective impression score. The abstract says visibility improved by up to 40 percent.

Plain presentation changes also helped. Making the text more fluent and easier to understand each gave a boost of roughly 15 to 30 percent.

On Perplexity, quotation addition improved the position-adjusted word count by about 22 percent, and statistics addition improved the subjective impression score by up to 37 percent.

What did not help

• Keyword stuffing. Repeating query keywords scored below the unchanged page: 17.8 against a baseline of 19.5 on the benchmark, and about 10 percent worse on Perplexity. The authors conclude that classic SEO tactics don’t carry over well.
• A more persuasive, authoritative tone. They found no significant improvement and write that these engines look fairly robust to it.

Which method worked depended on the topic

The paper says effectiveness varied by domain. Citing sources worked best on factual questions, and statistics helped most in law, government and opinion queries. So there is no single rewrite that works everywhere.

Limits you should know about

• It tested two generative engines. The authors say methods may need to change as these engines evolve.
• Part of the scoring came from GPT-3.5 applying a rubric, not from real readers.
• The authors did not measure effects on ordinary search rankings.
• It measures how much a source is used inside an answer. It does not measure clicks, visits or sales.
• We know of no study showing these changes raise sales for a small store, and this note does not claim one.

What Google says about its own AI features

Google’s page on AI features says a page “must be indexed and eligible to be shown in Google Search with a snippet”, and that there are “no additional requirements” or special optimizations to appear in AI Overviews or AI Mode. It recommends a good page experience, structured data that matches the visible text, and important content in text form. It also says these features count toward your overall search traffic in Search Console’s Performance report. The paper and Google’s page do not conflict: Google describes the entry ticket, the paper describes how sources compared once they were in.

What a store owner can do this week

1. Pick your five most-viewed product or collection pages. Where a number is true, state it with its source. Examples are material, size, warranty length and shipping time.
2. Where a customer or an expert has said something useful, quote them by name, with permission.
3. Link to the source for any claim you did not make yourself, such as a care standard or a test result.
4. Read each page aloud. Cut padding and long sentences. The study found plain text helped.
5. Do not add repeated keywords. The study found it hurt.
6. In Search Console, confirm that these pages are indexed and not set to nosnippet.
7. Change one page at a time and note the date, so you can look back. This is the same discipline as A/B testing, with far fewer data points.