Thoughts
Your website now has two audiences: the person who visits and the AI that decides whether to recommend you
A few months ago I noticed something new in the analytics for ventbusting.com. Referral traffic from ChatGPT. Not a flood. Enough to keep showing up. Someone asked an AI where to get their dryer vent cleaned in the Triad, and the model pointed them to Vent Busters.
That wasn’t happening three years ago. It is now. I don’t think most service business owners have caught up to what that means.
Traditional search is becoming optional
The numbers moved fast. ChatGPT referral traffic to the broader web grew 206% in 2025. Across ChatGPT, Perplexity, Claude, Gemini, and the rest of the category, AI referral traffic was up 357% year over year. Over 60% of search interactions now involve an AI-generated component. A 2025 Capgemini survey found 58% of users have already replaced traditional search engines with AI for product and service discovery.
The UX shift is simple. Instead of searching, reading ten results, and clicking around, someone asks a question in plain language and gets a synthesized answer with maybe one or two sources. The click happens after the AI has already made a recommendation.
For a service business, that changes how discovery works. You’re not only asking whether your site ranks for a keyword. You’re asking whether your business shows up when someone asks an AI who to call for dryer vent cleaning in Greensboro or Winston-Salem.
How AI systems pick what to cite
Search engines index keywords. AI systems read pages, pull out facts, and decide whether the content is clear and specific enough to quote in an answer.
Keyword stuffing doesn’t help. An AI reads your page the way a sharp customer would and decides if there’s anything worth citing: specific claims, clear service areas, named proof, reviews, and pages that actually answer the question. Vague marketing copy gets skipped.
For Vent Busters, the things probably driving AI referrals are the same things that made the Astro rebuild worth doing: fast pages, clean HTML, real customer reviews on the page, LocalBusiness schema with service area data, and content written around questions people actually search. The model can read the page, find the facts, and make a confident recommendation.
Schema markup and answer-shaped content
Most local businesses I look at are missing one of two things.
Structured data. Schema markup labels what your content means, not just what it says. LocalBusiness schema gives an AI your business name, service area, hours, and category without guessing. Review schema ties those five-star ratings to this specific business.
Schema has been local SEO advice for years. It matters more now because AI systems treat it as a trust signal. Two businesses in the same area with similar copy? The one with proper schema is easier to cite.
Content that answers a question directly. AI systems look for the shape of an answer, not the shape of a sales page. “Dryer vent cleaning removes lint buildup that causes fires and cuts dryer efficiency. Vent Busters services the Triad region including Greensboro, Winston-Salem, and High Point, with 15,000+ dryer vents serviced since 2017” is useful to an AI. “We are the leading dryer vent experts in the area” is a claim with nothing behind it.
A model can cite the number, the region, and the year. “Leading experts” isn’t a fact.
llms.txt
Worth knowing about: llms.txt. Proposed standard, sits at the root of your site (/llms.txt), plain markdown that tells AI systems what your site is about, which pages matter, and how to read your content.
Simple format. Site name in the header, short description, organized sections linking to key pages or their markdown equivalents. A curated map instead of forcing a crawler to infer structure from HTML.
Adoption is still spotty. GPTBot and ClaudeBot mostly crawl HTML directly and don’t consistently read llms.txt yet. The direction is clear though. On Astro, a root /llms.txt route takes about twenty minutes. Low effort to be early. Ignoring it costs you nothing today and might cost you something in a year.
What I’d actually do
If you’ve already built a decent website, you’re not starting from zero. Fast pages, useful content, honest reviews, and clean HTML all carry over into AI discoverability. Thin, slow, keyword-stuffed sites are the ones that’ll struggle. If that sounds like yours, it’s fixable. I help service businesses with this kind of work.
AI raises the bar for “real content” though. A page that exists only to rank for a location keyword doesn’t help a model trying to recommend a business. A page that says what you do, where you operate, what it costs, what customers have said, and what makes the business specific is content an AI can work with.
On Vent Busters I’m doing the boring work first: keep schema accurate, write better answers to the questions people ask about dryer vents, and get llms.txt in place before every local business is expected to have one.
Referrals with intent
AI referral traffic converts at 7.1% according to 2026 benchmarks, just below paid search at 7.8%. Above organic search, social, email, and direct. Fewer visits than traditional search right now, but the people arriving from AI recommendations arrive with intent. They asked a question, got an answer, clicked the business the model named.
Different user than someone browsing ten blue links. The AI already did the comparison. By the time they land on your site, you’re the recommendation.
That’s what people call GEO, generative engine optimization. Newer and less settled than traditional SEO. The logic is familiar: fast site, specific copy, proof from real customers, content that’s easy to parse.
Two audiences now. Same fundamentals as before.
The Vent Busters website runs on Astro with LocalBusiness schema, Umami analytics, and SerpBear for keyword tracking. Read the ventbusting.com case study for the full rebuild story. I also wrote about making website content easier for AI agents to read.