AI Search Optimisation: Get Cited, Not Just Ranked
When customers ask an AI directly, the answer often arrives before they click anything. The question shifts from “how do I rank higher” to “when the AI answers, does it cite me”.
How AI search differs from Google ranking
Google ranking orders links; AI search generates an answer. The first decides which page is top, the second decides which content gets extracted, synthesised, and cited.
An AI reads several sources, extracts the passages it judges credible, clear, and structured, and synthesises a reply — sometimes with citations. So the emphasis moves from page authority to whether content is easy for a machine to read and verify.
Neither replaces the other. Content an AI cites still has to be indexed and understood first — SEO is the foundation, and AI search optimisation is built on it.
Two terms, one workstream
AEO
AEO (Answer Engine Optimization) is shaping content so a machine can extract it directly as an answer: explicit questions, concise direct answers, self-contained passages, and matching structured data.
- Headings phrased as questions, with the answer landing in the first two or three sentences
- Structured FAQs and answer blocks paired with FAQPage schema
- Comparisons set out as tables or lists so they extract cleanly
- No vague “it depends” — state the range first, then what affects it
GEO
GEO (Generative Engine Optimization) is building the signals that make generative AI willing to trust and cite you: named authors and professional review, checkable first-hand evidence, and organisation details that stay consistent across platforms.
- E-E-A-T: named authors, credentials on show, professional reviewer markup
- Entity consistency: name, address, phone, and services identical across your site, GBP, and directories
- First-hand evidence: real figures with their measurement period, source, and method
- Off-site mentions and knowledge-graph signals
AEO and GEO overlap heavily — two sides of one workstream, one handling format and the other credibility. They share the same content and technical base, not two budgets.
How we work
1. Establish a prompt baseline
A fixed set of questions across brand, category, problem, comparison, location, and buying stage. Fixing them is what makes a trend readable — AI answers vary by phrasing and over time, so a single result means little.
2. Citation map
Record which URLs and domains each question cites, which brands appear, and what the cited content has in common. The competitor pages that get cited are usually your content gap.
3. Eligibility check
Confirm the content can actually be read: crawlable, indexable, present in rendered text (not only after JavaScript runs), with correct canonicals and hreflang. Fail this step and nothing downstream matters.
4. Content upgrade
Add first-hand evidence, direct answers, comparison tables, method notes, authorship, and why it was updated. The goal isn't more words — it's verifiable ones.
5. Entity and off-site consistency
Organization / Person structured data, GBP and directory details, credible third-party mentions — so an AI can confirm these references all point to one entity.
6. Monitoring
Re-run the same prompt set monthly, recording mention rate, cited sources, factual accuracy, and source diversity — read alongside impression changes in Search Console.
AI engine coverage
Where our AEO / GEO work is directed
Customers no longer search on Google alone. We optimise content structure, answer formatting, and authority signals for the AI search and answer engines below, working toward being referenced when customers ask — whether any engine cites you is decided by its model and cannot be guaranteed.
ChatGPT
OpenAI's assistant — among the most widely used AI tools.
Google AI Overviews
The AI summary atop Google results, directly affecting organic visibility.
Gemini
Google's generative assistant, integrated across Search and Workspace.
Perplexity
An AI search tool that footnotes its sources and favours fresh pages.
DeepSeek
A model strong in Chinese-language contexts, increasingly used by Chinese-speaking users.
Claude
Anthropic's assistant, often used for research and long-form reading.
Doubao
ByteDance's assistant, widely used in mainland China and by some Chinese-speaking users.
All names and marks are trademarks of their respective owners, shown for identification only. They do not imply partnership, sponsorship, or endorsement.
Honestly: what doesn't work
This category is full of overstated claims. These are worth saying plainly, even though they make the service sound less magical.
llms.txt won't get you cited in Google's AI features
Google's AI-optimisation guidance of 15 May 2026 states plainly that a site needs no llms.txt or AI-specific file to appear in Search — including AI Overviews and AI Mode — because Google Search doesn't use it. An Ahrefs study across 137,000 sites found roughly 97% of llms.txt files were never fetched. We keep one on our own site for brand consistency, but we won't sell it to you as a visibility mechanism.
There is no “AI-specific schema”
Structured data helps, but it uses existing standards (Organization, FAQPage, Article, and so on) and must match visible content. Claims of special AI markup that accelerates citation don't hold up.
You don't need a page per query variation
Mass near-duplicate pages covering query variations is an old-SEO habit. It does nothing for AI citation and dilutes your topic while wasting crawl budget.
No one can guarantee an AI citation
Whether you're cited is decided by each platform's model. Any service promising placement in AI answers isn't credible. What we can do is improve the conditions: crawlable, understandable, evidenced, entity-consistent, and credibly sourced.
How it's measured
There's no official report for AI citations, so the measurement method itself has to be transparent. We re-run a fixed prompt set monthly and record four things: mention rate (how many questions surface you), the source URLs cited, the factual accuracy of what's said, and source diversity.
We read that against Search Console: rising impressions with a falling click-through rate is usually the signature of answer-style search rather than failing content. Reading both together avoids drawing the wrong conclusion.
We report it as it is, including months with no progress. This data is volatile, and one month is not a verdict.
Frequently Asked Questions
Can you guarantee we get cited by ChatGPT or AI Overviews?
Does adding llms.txt make AI find us?
Is AI search optimisation charged separately?
If AI answers directly, will our traffic fall?
How long before AI citations change?
Is it worth doing on a site with little content?
Want to know what AI currently says about your brand?
The free search-visibility review includes testing your brand and category questions in AI tools — whether you're mentioned, and who gets cited instead.