Web & technologies
SEO & GEO
Found in both places people now search — ranking in Google's results, and cited by name inside ChatGPT, Perplexity and AI Overviews, where a click is no longer guaranteed.
The short version
Search now has two exits. One is the blue link you have always optimised for. The other is an answer written by a model that read your page and may or may not name you. Those two goals overlap more than people expect, and the real difference is in how the answers are written and how you measure a mention that produces no click at all.
What is included
- A technical crawl of the whole site, with findings ranked by what is costing you traffic rather than by how many red rows the tool managed to produce.
- Fixes shipped rather than merely listed — indexation, canonicals, redirect chains, internal links, Core Web Vitals, and the templates duplicating your titles.
- A keyword and question map grouped by what the searcher actually wants, separating the terms that deserve a page from the ones that deserve a paragraph.
- Pages restructured so one section answers one question in its own words, because a passage is the unit a model retrieves and quotes.
- Schema markup that matches the page's real content — Organization, Article, FAQPage, Product, LocalBusiness — plus an llms.txt file and crawlable HTML for AI crawlers.
- Monthly reporting that puts Search Console clicks next to prompt testing across ChatGPT, Perplexity and Google AI Overviews for the questions you want to own.
How we work
Our approach
The stages this work runs through, in order. Every one of them ends in something you can see or sign off.
-
Step 01
Fix the foundations first
There is little point writing new pages while the existing ones can't be crawled properly. So we start with whatever stops indexing or dilutes it: redirect chains, thin duplicate templates, filter URLs multiplying into thousands, missing or contradictory canonicals, slow pages on mobile. This work is unglamorous and it is where we insist on starting, even if you came to us wanting new content. You get the list, we make the changes with your sign-off, then recheck in Search Console rather than assuming the fix landed.
-
Step 02
Map questions, not keywords
A keyword list tells you what people typed. It doesn't tell you what they wanted, and wants are what models answer. We group terms into the questions sitting behind them, decide which deserve a page of their own and which are a section, and mark the ones your site already half-answers. That last group is where movement comes fastest, because improving a page already sitting on page two beats writing a new page from nothing.
-
Step 03
Write to be quoted
A model retrieves passages, not whole pages. So each section takes a plain question as its heading and answers it inside the first two sentences, with the specifics — the number, the condition, the exception — in that same passage rather than three paragraphs away. Vague copy can't be quoted, so it isn't. We also make your organisation, what it does and where it operates consistent across the site and your public profiles, because that consistency is how a model decides you are one entity.
-
Step 04
Measure the mentions too
Clicks stay the primary number, taken from Search Console and split into branded and non-branded so growth isn't just your own name coming back. Alongside that we run a fixed set of prompts across the main assistants each month, record whether you are named and what is said about you, and keep the prompt wording constant so the comparison means something. It is a rough instrument. It is also the only honest way to watch the AI side move.
Before you ask
Questions we get asked first
How long until we see movement?
Technical fixes can show in Search Console within 2 to 4 weeks, because indexing responds quickly once you stop obstructing it. Movement on competitive non-branded terms is a 3 to 6 month question, and anyone promising faster is guessing. Citations inside AI assistants move on their own schedule, because each one refreshes its sources differently and none of them publish the rules. We report monthly from the first month, so you can see direction well before you see the result.
How is it priced?
Two ways, and we will tell you which one fits. A one-off audit and fix programme is priced as one job in pounds after we have crawled the site, which suits a site that has never had the technical work done properly. Ongoing work is a monthly fee for an agreed number of days, with that month's plan sent in advance and the work logged. No percentage of traffic and no rank-based bonuses, because those push agencies towards easy terms nobody searches for.
What do you need from us?
Search Console and analytics access on day one, and server log access if you have it, since logs show what crawlers actually did rather than what a tool inferred. Either permission to publish changes ourselves, or a developer we can hand precise instructions to. Also someone who knows the subject well enough to confirm a rewritten page is still true, because the quickest way to lose ground is to publish confident nonsense in your own field.
How do you measure GEO when there are no clicks to count?
By testing prompts and counting mentions. We agree 20 to 30 questions a buyer might genuinely put to an assistant, run them monthly in identical wording, and record whether you are named, what is said, and which page is cited. We watch referral traffic from assistant domains in analytics, which is small but real. And we track branded search volume, because a mention with no link often reappears later as somebody typing your name into Google.
Want this scoped for your situation?
Tell us the goal and the deadline. You get an honest read on whether this is the right service, a fixed price if it is, and a straight no if it isn't.
Or email [email protected] directly. We reply within one working day.