AI search optimisation
Being retrievable, and being quoted
Two separate problems. Whether an AI system can reach your pages and read them at all, and whether what it finds there is worth quoting. The first is mechanical. The second is editorial, and much harder.
Two problems
The difference
Think of it as getting into the club versus being handed the microphone.
Fetchability is purely mechanical. Can an AI agent reach your server, open the door, and grab the text without choking on broken scripts or blocked paths?
Citability is editorial and reputational. Once the model has read the page, is the information framed cleanly enough that it takes a bite-sized piece, quotes it, and points back to you as the source?
Substance
What it covers
Fetchability
First, the guest list — robots.txt and firewall settings. A surprising number of teams accidentally block GPTBot, PerplexityBot or ClaudeBot, or have Cloudflare settings that aggressively challenge them.
Then plain-text delivery. A lot of modern sites hide their best content behind heavy client-side JavaScript. If a bot fetches the page and gets a blank white shell because it doesn’t spend the compute to render your scripts, you don’t exist.
Clean HTML, delivered fast, with no crawl barriers.
Citability
Models don’t think in 3,000-word blog posts. They think in discrete chunks. A quotable passage has three things.
Definition-first formatting
Answering “what is X” in the first sentence, rather than burying it under five paragraphs of introduction.
Concrete data over adjectives
“Cut load times by 42%” instead of “significantly improved performance.” Models favour factual, falsifiable statements.
Unambiguous context
A paragraph that makes complete sense on its own, without needing the rest of the page to explain what “it” refers to.
The test: if an AI can lift two sentences directly into its answer without sounding awkward, you’ve built a citable asset.
Measurement
How it gets measured
There’s no clean rank tracker for AI answers yet, so the work starts with a prompt benchmark.
I take the top twenty to fifty conversational buyer queries your audience actually asks Perplexity, ChatGPT or Google’s AI Overviews, and run them at baseline. Are you mentioned? Are you cited as a source link? Is the answer accurate, or is the model hallucinating about you?
Progress looks like your brand appearing as a recommended provider or cited source across that prompt set, paired with referral traffic from AI platforms in your analytics.
This benchmark runs on Kreametrix itself. I built it for this practice before selling it to anyone.
Honesty
What I can’t promise
That you’ll appear in every generative answer, or that any placement is permanent.
Large language models are probabilistic, non-deterministic engines. They don’t produce the same text twice for the same prompt. And platforms push model updates that rewrite how they source links overnight.
What I can promise is removing every technical roadblock, and engineering your content so it’s the easiest and most authoritative answer available for a model to reach for.
Your side
What I need from you
Access to the technical infrastructure — CMS, hosting, CDN and firewall. And access to your subject matter experts.
I can structure the data and clear the bot paths. I can’t invent proprietary benchmarks, unique company data, or real industry insight out of thin air. Citability runs on substance. If your team can’t give me genuine, authoritative facts to package, the model will cite the competitor who can.
Fit
When this is the wrong thing to buy
If the business doesn’t have a clear point of view, proprietary data, or basic market trust.
A generic affiliate site, a drop-shipping store, or a content operation rewriting what’s already on Wikipedia won’t be saved by AI optimisation. Models summarise the consensus. If you bring no original value to that consensus, there’s nothing there to cite.
Engagement
How it’s scoped
The technical cleanup is largely a one-off project — auditing crawler access, fixing server bottlenecks, patching JavaScript delivery, setting the foundation.
Citability is ongoing. As competitors publish, as prompt behaviour changes, and as new platforms launch, staying the cited authority means consistently feeding new structured material into the ecosystem.
Related services