In Q1 2026 we ran a focused 12-week experiment to get one client - a Newcastle plumber - cited by ChatGPT and Perplexity for the query "plumber near me in Newcastle" and its variants. We started from zero citations. By week 10 he was being cited in roughly 40% of ChatGPT responses and 65% of Perplexity responses. This is the playbook.
The prompt-test methodology
Before doing anything we built a standard prompt log: 12 variants of "plumber near me" queries that a Newcastle homeowner might actually ask. We ran the full set against ChatGPT (with web search on) and Perplexity weekly, logging which businesses were cited. Baseline citation rate for our client: 0%.
You cannot improve what you don't measure. The first deliverable on any AI-search engagement should be the prompt log. Without it you're guessing.
Weeks 1-3: schema overhaul
We rebuilt the site's structured data. Switched from generic LocalBusiness to Plumber (see the schema field note). Added areaServed as an array of GeoCircle objects for each suburb. Added hasOfferCatalog with every service and a "from" price. Added knowsAbout as an array of 14 specific services. Added openingHoursSpecification as a day-by-day array.
By week 3, Perplexity citations went from 0% to 15% on the prompt log. ChatGPT didn't move yet. The schema work was necessary but not sufficient.
Weeks 4-7: third-party mentions
ChatGPT's web search lean is heavier on third-party trust signals than on first-party schema. We got the client mentioned in:
- Two local-business roundup posts on Newcastle-focused community sites (genuine outreach, not paid placements).
- The local Federation of Master Builders listing (free, took 20 minutes, often overlooked).
- A trade-association directory specific to plumbing.
- A guest-author post on a regional home-improvement blog with a contextual link back.
By week 7, ChatGPT citations had climbed to 25%. Perplexity was at 45%. The third-party signal was the unlock.
Weeks 8-10: content depth
We added five long-form, question-shaped pages to the local web design build, each answering a specific homeowner question with a structured answer (H2 = question, first paragraph = direct answer, rest of the page = depth). Topics: "Why is my combi boiler losing pressure?", "How much does a new bathroom cost in Newcastle?", "What's the difference between a Gas Safe and a non-registered plumber?", etc.
This is the kind of content AI search loves: directly answers a question, has clear structure, has the trust signal of being on a properly-schema'd business site. By week 10, ChatGPT citations hit 40%, Perplexity hit 65%.
What didn't work
Things we tried that didn't move the needle:
- Stuffing FAQPage schema everywhere. Some lift; less than we hoped.
- Submitting to ChatGPT's "Suggest a website" form. Couldn't measure any effect.
- Buying directory listings on the long-tail "100 free directories" list. Zero detectable lift, and a citation-cleanup headache.
- Adding our own llms.txt to the site (it's a great signal, but we couldn't isolate its specific contribution - the third-party work happened in the same period).
The bigger picture
AI search citation in 2026 is the new map pack. The mechanics are different but the principle is identical: trust signals compound. Schema, citations, third-party mentions and content depth all reinforce each other. There is no single hack.
The good news for local web design for tradespeople is that local businesses have a structural advantage in AI search vs national brands. AI search wants to give a specific recommendation, and "the local Gas Safe plumber in Heaton with 200+ reviews and 12 years trading" is a much better recommendation than "British Gas". Specificity wins.
The Perplexity referral data
After the 12 weeks, Perplexity referrals were 2.1% of total site traffic. Small but rising. More interestingly, the conversion rate of those visits was 14% vs the site's overall 7.4%. AI-search traffic comes pre-qualified; the user has already had the LLM endorse the business before they click.
The next 12 weeks
We're now running the same experiment on three more clients in different trades to see if the pattern holds. Early signs: yes, with the third-party-mentions piece being the dominant lever for ChatGPT specifically. We'll publish the follow-up when the data lands.
Every Built Local plan now includes a quarterly AI-search audit and prompt-test cycle as part of the £49/month. See the AI search pillar for the broader strategy, or read the pillar guide for where AI search fits into the full UK local web design picture.