Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

GEO: Generative Engine Optimization

Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan et al. — 2024 — ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2024)

conference paper  ·   ·  deep dive

Key findings

• When a source was already in the AI's pool of documents to read, certain content edits made it much more likely to get cited in the AI's final answer — the best interventions increased citation visibility by up to 40%.
• Adding statistics and direct quotes from credible sources worked best for factual questions (e.g., 'What causes inflation?'), while using confident, authoritative language worked better for opinion-style questions (e.g., 'What is the best diet?').
• Not all tricks worked the same everywhere — the effectiveness of each strategy varied significantly by topic, meaning there is no single universal approach to GEO.
• Tests on the real-world AI search engine Perplexity.ai showed similar results, with visibility improvements of up to 37%, suggesting the lab findings have some real-world relevance — though the same retrieval-stage limitation applies.

Marketing implications

• If you write content for a company website, start adding real statistics with source citations and direct quotes from credible experts — this makes your page more likely to get mentioned in AI-generated answers on tools like Perplexity.ai or Bing Copilot.
• Don't use the same approach for every article. For how-to or factual content, lean heavily on numbers and data. For opinion or recommendation content (e.g., 'best tools for X'), use direct, confident language.
• Treat GEO as a new layer on top of traditional SEO, not a replacement — you still need your page to get crawled and retrieved before these content tricks can help.

Paper B

Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026)

Martinez — 2026 — arXiv (preprint)

preprint  ·   ·  read now

Key findings

• GEO is not one simple problem — it actually involves at least 11 different stages, from whether an AI engine even activates a search, all the way through to whether a user acts on what the AI told them. Most vendors and practitioners treat it like a single ranking problem, which this survey says is wrong.
• There is some evidence that when your content is retrieved (pulled into the AI's context window), it can genuinely influence whether the AI cites or uses it. But this only holds in controlled lab-style settings — not in the real world over time.
• No GEO technique reviewed has yet shown it can reliably and consistently improve how often a brand gets mentioned by AI engines across different platforms over time — or that it drives more web traffic, engagement, or sales as a result.
• The paper explicitly warns that vendor claims about GEO tools and services are not backed by solid evidence. Marketers buying these tools should be skeptical until independent research proves otherwise.

Marketing implications

• If a vendor is pitching you a GEO tool that promises to get your brand mentioned more by ChatGPT or Perplexity, ask them for peer-reviewed proof that it works at scale over time. This survey says that proof does not exist yet — so push back hard before signing a contract.
• Focus your GEO efforts on getting your content retrieved first (making sure AI engines can actually find and pull in your content), since that retrieval step is where the most evidence of influence exists — not on tricks to game the final answer.
• Treat any GEO program you run right now as an experiment: set up measurement before you start, track whether AI mentions actually translate to web traffic or sales, and don't assume any improvement in AI citations is driving business results without the data to back it up.

Paper C

What Generative Search Engines Like and How to Optimize Web Content Cooperatively

Wu — 2025 — arXiv / OpenReview (preprint)

preprint  ·   ·  test this week

Key findings

• AutoGEO's automated content rewrites improved how often pages appeared in AI-generated answers by an average of 35.99% across the tested search engines, while keeping the content accurate and easy to read.
• The rules AutoGEO learned for one type of content did NOT work equally well for other types. Rules learned from general web content overlapped a lot with other general content, but barely transferred to e-commerce product pages — meaning one universal 'optimize for AI search' checklist is not enough.
• Different AI search engines (Gemini, GPT, Claude) appear to have different preferences for what content they cite, reinforcing that engine-specific optimization matters.
• The system can operate automatically, suggesting AI-driven content rewriting tools could potentially replace or reduce the need for humans to manually optimize content for generative search.

Marketing implications

• If you manage content for e-commerce product pages, don't assume what works for your blog or general web content will also make your product pages show up in AI search results — you need separate optimization rules for each.
• If you're an SEO consultant or agency, start treating Gemini Search, ChatGPT search, and Claude differently rather than applying one unified GEO checklist — this paper suggests each engine has its own citation preferences.
• Watch this space: automated tools that rewrite your content to match what AI search engines prefer are coming. If AutoGEO's results hold up, this could become a standard part of content workflow — worth piloting now before competitors do.

Apple Podcasts  ·  Spotify  ·  Buzzsprout