How to optimise a site for AI engines

Step 1 — Baseline: measure before optimising

Before any action, take an honest snapshot of your site's current visibility across AI surfaces. Without a baseline, no iteration is credible.

1. Pick 10 to 20 strategic queries covering top, middle and bottom of funnel.
2. Run each on ChatGPT, Perplexity, Gemini, Claude and Google (AI Overviews).
3. Note for each query: sources cited, presence or absence of your brand, quality of the answer (accurate, biased, outdated).
4. Log results in a dated table.

Complement with Google Search Console (AI Overviews report when available), and, if possible, a monitoring tool (Profound, Peec AI, Otterly, Scrunch, AthenaHQ).

Step 2 — Structure content for retrieval

AI engines split documents into chunks and index passages. Poorly structured content becomes invisible, even if good. Operational rules:

See the dedicated page on content structure for LLMs.

Step 3 — Consolidate semantic authority

LLMs reason in entities. To be cited on a topic, your site must be recognised as a legitimate source on that topic. Levers:

• Topic clusters: one pillar, 5 to 10 satellites, strong internal linking.
• Entity co-occurrence: pages cite domain markers (competitors, tools, methods) with nuance.
• Disambiguation: your brand is tied to its expertise consistently on- and off-site (Wikipedia, social profiles, directories).
• Inbound links from topically aligned sites. One relevant link beats ten generic ones.
• schema.org: Organization, Article, FAQPage, BreadcrumbList properly formed.

Step 4 — Open the site to AI bots (or arbitrate) and reinforce the technical layer

Crawl is a necessary condition. Check:

Step 5 — Make content citable

A page can be well-indexed and never cited if it offers no extractable claim. LLMs prefer passages that stand alone and carry clear informational value.

1. Sharp statements. Avoid "it may appear that", "some think that".
2. Dated and sourced figures. A figure without a source is suspicious; a figure without a date ages badly.
3. Primary-source citations. LLMs reward pages that cite other credible pages.
4. Repeatable formats. Answer a question with a firm paragraph, followed by an example and a nuance.
5. Identified author. A page with well-filled Organization and/or Person schema.org reinforces trust.

Step 6 — Measure, iterate, maintain

AI optimisation is a long-run discipline, not a sprint. Recommended cadence:

Where to start, concretely

If you can only do one thing: audit a pillar page (your strongest page on a strategic topic) with the LOOP checklist, fix it in half a day, and re-test on AI engines 3 weeks later. That's the test with the highest learning per hour invested.