The research paper: GEO as content science
Generative engine optimisation entered public conversation through an academic paper: GEO: Generative Engine Optimisation (Aggarwal et al., arXiv 2023–2024).
The researchers asked: as answers shift from "ten blue links" to AI-composed responses, how should websites change to get cited inside those answers?
They tested content tactics — adding statistics, credible quotations, simpler phrasing, structures AI can quote accurately. In their benchmark, some tactics measurably increased visibility in AI-generated responses.
What the paper is not:
- A guide to robots.txt or OpenAI's crawler policies
- A llms.txt specification
- A product review of Semrush or Ahrefs
- Proof that any specific website will appear in ChatGPT
The paper studies what you write. Many owners need to fix whether OpenAI, Anthropic, or Perplexity's crawlers can read you before content tactics matter.
Example: A consultancy rewrites service pages with statistics and expert quotes — straight from paper-inspired tactics. Google rankings hold per Search Console. ChatGPT still names competitors. A separate check shows OpenAI's OAI-SearchBot blocked — often after enabling Cloudflare's Block AI bots rule, which Cloudflare documents blocks GPTBot, ClaudeBot, and other crawlers site-wide. Content work on an unreadable site doesn't help.
Commercial GEO: three product camps
After the paper, vendors adopted "GEO" fast. In the market today, the acronym usually means one of three different jobs — each with established tools and DIY paths:
All three are legitimate. None replaces the others automatically.
- Content GEO won't unblock OAI-SearchBot if Cloudflare returns 403.
- Monitoring GEO (Otterly tracking whether Perplexity cites you) won't generate a corrected robots.txt.
- Infrastructure GEO won't rewrite your blog posts or track competitive mention share over time.
When a vendor says "you need GEO," ask which row they mean.
Content GEO example: A team uses Clearscope to optimise articles for extractability — stats up front, clearer headings. Useful once crawlers can reach the pages.
Monitoring GEO example: A brand subscribes to Profound to track how often ChatGPT and Perplexity name them vs. competitors on a set of buyer prompts. Useful when access is already confirmed.
Infrastructure GEO example: An owner discovers PerplexityBot is allowed but llms.txt is missing. Options include handing robots.txt to a developer using OpenAI's, Anthropic's, and Google's crawler docs, reviewing Cloudflare bot settings, or using a readiness tool that generates files plus CMS steps.
None of this contradicts the category split between SEO platforms and GEO tools — it clarifies which GEO layer sits on the AI side.
Which layer to fix first
A practical sequence most consultants and vendors agree on — regardless of which product you pick:
- Access — Can OpenAI's, Anthropic's, and Perplexity's crawlers reach the site? (AI crawlers SEO tools don't check)
- Structure — llms.txt, schema.org markup, clear service facts machines can parse
- Content — Paper-style tactics: stats, quotes, extractable copy (MarketMuse/Clearscope territory)
- Authority — Reviews, directories, third-party mentions
- Monitoring — Profound, Otterly, or Peec AI once the foundation is sound
Skipping to step 3 or 5 while step 1 is broken is the most common waste we see. Ahrefs and Semrush can report green on Googlebot while step 1 fails for OAI-SearchBot.
If the problem sounds like "how much pipeline am I losing?" start with how buyers use ChatGPT to find vendors. If competitors are cited while you rank higher on Google, see why ChatGPT recommends competitors.
Honest limit: no GEO layer — academic or commercial — guarantees ChatGPT placement. OpenAI, Google, and Perplexity control what their systems cite.
GEO vs SEO vs AEO (thirty seconds)
For why Google rankings don't transfer to ChatGPT automatically, see Chat GPT SEO Google.