Get found by AI.
Make your business visible in the answers your customers are already asking for. We improve the structure, content, technical signals and authority that AI systems use to understand and recommend businesses — on the site you already have, not a rebuild you didn't ask for.
Microsoft now measures what we optimise for.
Bing Webmaster Tools added AI Performance metrics in February 2026 — Total Citations, Average Cited Pages, Grounding Queries. Cloudflare's AI Crawl Control reports the adjacent crawl-level signal. Google still folds AI Overviews into standard Search Console with no dedicated dashboard yet. The signal moved from anecdotal to measurable in months — being cited by AI is now a number you can report on. We've been building for this since before the dashboards existed.
Read Microsoft's announcement →Six layers, and what each one costs you.
Six categories an answer engine checks. Each one fails closed by default — unless it has been explicitly addressed, the gap is there. Most sites we audit fail four or five without knowing it.
AI crawler permissions
Most sites quietly block the AI crawlers without realising — either by a default robots.txt that was never revisited, or by bot protection doing its job too well. Without explicit permission for the major foundation-model agents, you are invisible before the question is even asked.
- Access audit across the major model crawlers
- robots.txt directive review
- Bot-protection and WAF reconciliation
- Per-agent allow and deny policy
- Post-deploy access verification
llms.txt at the root
An emerging standard for guiding language models through a site. It is the highest-impact file missing on almost everything we audit. Not yet universal — but Anthropic, OpenAI and Mistral have signalled support, and the sites that adopted early are the ones already being read properly.
- Root-level llms.txt authoring
- Site-structure summary for model traversal
- Priority-path definition
- Spec tracking as the standard settles
- Revalidation when content changes
Structured data across content types
AI systems extract from sites that hand them machine-readable summaries. Most sites hand them nothing, so the model has to infer what the business does from prose — and increasingly it simply moves on to a source that made it easy.
- Schema coverage mapped across page types
- Organisation and LocalBusiness markup
- Service, FAQ and Article types
- Validation against Schema.org and Google
- Cross-page graph consistency
Rendering layer
AI crawlers read the HTML your server returns, not what JavaScript assembles a moment later. Sites built as client-side single-page apps are close to blank to them. This is not a fragile edge case — it is the foundation everything else sits on.
- Server-rendered HTML verification
- Identification of client-only content
- Crawler-eye HTML diffing
- Render-path remediation scoping
- Core Web Vitals impact check
Content shaped for citation
Fact-dense, cleanly structured, with clear authorship and credibility signals. This is a different discipline from SEO writing. Search engines rank pages; answer engines look for a source they trust enough to quote by name.
- Answer-first structural pass
- Fact density and specificity review
- Authorship and credibility signals
- Question-to-answer mapping
- Quotable-block formatting
Entity identity
Without machine-readable identity markers, an answer engine cannot reliably attribute what it quotes back to your business. A citation nobody can trace to you is a wasted impression.
- Identifier and sameAs markup
- Name, address and phone consistency
- Knowledge-panel alignment
- Brand-to-entity disambiguation
- Attribution spot-checks in live AI answers
Scoped against the actual gap.
Most optimisation passes are scoped against the audit. We don't quote in the dark — we run the audit, then quote against the specific gaps.
- Free Dyfyn AI checker. Fifteen seconds, a score out of 100 across the five visible categories. On this site. Tells us both where you sit and what's worth fixing first.
- Detailed audit report (£350). The line-by-line breakdown — what each finding means commercially, prioritised, with effort estimates. Credited against any subsequent engagement.
- Optimisation pass. Scoped to the gaps the audit surfaces. Quote-based after the discovery call. Wider scope means a rebuild's probably the better answer.
- Care plan. Optimisation isn't one-and-done. Standards evolve; competitors update; the care plan keeps you current.
A Cardiff studio, deliberately small.
No account manager, no junior doing the actual work, no discovery theatre. You talk to the person who runs the optimisation.
Dyfyn is run from Cardiff by Ed Jones, twenty-five years into building for the web. The studio stays small on purpose and leans hard on AI tooling to do the work a team of ten would have done a decade ago — a slightly unusual thing to admit, except that it is the same argument we are making to you about your own website.
The techniques on this page were not learned at a conference. They came off a portfolio of finance sites we own and pay for, where getting this wrong costs us directly. That is also why we'll tell you when the work isn't worth doing — there is no delivery team's calendar here that needs filling.
Related work.
Proof, tooling and positioning — the three things worth reading before you decide whether this is a conversation worth having.
Answer engine optimisation, asked plainly.
The questions that come up before anyone signs anything — including the ones where the honest answer is no.
SEO is the work of ranking in a list of ten blue links. Answer engine optimisation is the work of being the source an AI quotes when it replies in a sentence instead of a list. The two overlap, but they are not the same job — and a site can do well at one while failing the other.
We have watched it happen on our own portfolio: a finance site still on page six of Google, named in 51% of the answers we measure. Same site, same content, two different verdicts, because the two systems are judging different things.
The work targets the signals shared across ChatGPT, Claude, Grok and Perplexity, plus Google's AI Overviews. There is no per-engine switch to flip — they are reading broadly the same things about your site, which is why the readiness stack generalises.
Where they differ is in what you can measure afterwards. Bing Webmaster Tools now reports citations directly; Google still folds AI Overviews into standard Search Console with no dedicated dashboard.
Start with the free audit — fifteen seconds, a score out of 100 across the visible categories. That is usually enough to tell whether the problem is a handful of gaps or something structural.
If it warrants going deeper, the detailed report walks every finding line by line with what it costs you commercially. We would rather scope against a real diagnosis than guess at a proposal.
The audit is free and the detailed report is £350, credited against any work that follows. The optimisation pass itself is quoted against whatever the audit actually surfaces — we do not price in the dark.
If the audit shows the gaps run deeper than an optimisation pass can reach, we will say so. A rebuild is sometimes the honest answer, and it is cheaper to hear that early.
A focused pass on a site that is structurally sound is typically a matter of weeks, not months. The variable is almost always the rendering layer — sites needing work there take longer, because that is foundational rather than cosmetic.
You get a timeline agreed before anything starts.
No, and we would rather say so up front. If nobody is asking an AI assistant questions your business could answer, the work has nothing to attach to. Some sectors are years away from that; some are already there.
It matters most where people research before they buy or hire — trades, professional services, anything with a considered decision attached. The audit is free precisely so you can find out which side of that line you sit on without spending anything.
Usually, yes — that is the whole point of this service as opposed to a rebuild. Where the underlying stack is worth keeping, we layer the readiness work onto it and leave the site you already have in place.
The exception is the rendering layer. If the site is a client-side app that returns an empty shell to crawlers, no amount of layering fixes that, and we will tell you rather than take the work.
Yes. Every technique here runs on a portfolio of finance sites we own, which is where we found out what actually moves the needle and what only sounds like it should. The securedloanrates case study is one of them.
It also means we are not advising from theory. When we say a change is worth making, it is because we watched it work somewhere we carried the cost of being wrong.
Run the audit, see the gap.
Fastest way to know where you stand is fifteen seconds at the audit. After that we'll know whether optimisation or a rebuild fits.