Ask ChatGPT to recommend a plumber in your town. Seriously — open the app and try it before you read on.
You’ll get two or three names, maybe with phone numbers, and a sentence on why each made the cut. No results page. No ads to scroll past. No page two. If your business isn’t one of those names, that customer never knew you existed, and there’s no position eleven to claw back from.
I’m Subhash Makkena, founder of Techlancer. We’re a software firm in Naperville, Illinois — we’ve been building web applications since 2016, and our marketing practice grew out of that engineering work. AI search visibility is a problem we took seriously enough to implement on our own site first: JSON-LD structured data written by hand, an llms.txt file at the root, pages restructured so a machine can lift a clean answer out of them. This guide is what we learned doing that work, organized so you can execute most of it yourself.
How do AI assistants decide which businesses to recommend?
When you ask an assistant for a recommendation, it typically runs a live web search, reads the top results, cross-checks the facts it finds about each business, and names the ones it can verify and quote with confidence. It isn’t consulting a secret directory. It’s reading the same public web you can see — just faster, and with far less patience for ambiguity.
There are two paths into an answer, and they move at very different speeds.
Training data is the slow path. Whatever the model absorbed about your business before it shipped is baked in until the next model update, which you don’t control and can’t schedule. If you’re mentioned there, great. If not, there’s no form to fill out.
Retrieval is the fast path, and it’s the one worth optimizing. ChatGPT’s search mode, Perplexity, Gemini, and Claude with web search all fetch live pages at question time. Each retrieves a little differently — Gemini leans on Google’s index, Perplexity runs its own crawler, ChatGPT mixes Bing with its own bots — but the pattern is the same: search, read, verify, cite.
Watching what actually gets cited, five signals show up over and over:
- Crawl access. If GPTBot, ClaudeBot, or PerplexityBot can’t fetch your pages — blocked in robots.txt, hidden behind JavaScript rendering, or timing out on a slow server — you’re invisible before the contest starts.
- Extractable answers. Assistants quote pages that answer questions directly: a question-shaped heading with a straight two- or three-sentence answer underneath. A claim buried in paragraph six of a wall of text doesn’t get lifted.
- Entity consistency. The model cross-references your name, address, phone, and service claims across your site, your Google Business Profile, and directories. When the facts disagree, the model hedges. When it hedges, it names someone else.
- Corroboration. Reviews, local press, directory listings, and forum mentions all confirm you’re real and reputable. An assistant recommending a business is staking its credibility on you; it prefers businesses other sources vouch for.
- Existing search rank. Retrieval starts from a web index, so pages that already rank get read first. Your SEO foundation isn’t obsolete — it’s the front door.
None of this is exotic. It’s the discipline of being legible to machines, done thoroughly.
Is AI search visibility different from regular SEO?
It overlaps heavily — maybe seventy percent of the work is identical — but the emphasis shifts. Traditional SEO fights for positions on a results page a human will scan; AI search optimization aims to be the source an assistant quotes when it composes an answer. That rewards structured data, direct answers, and consistent facts more than raw link volume.
The practical difference shows up in how you write. An SEO-era service page opens with three paragraphs of throat-clearing before it says anything concrete. A page built for AI citation states the answer in the first two sentences under the heading, then elaborates. Humans prefer the second version too, which is why this isn’t a tax on your existing marketing — the same formatting wins featured snippets and Google’s AI Overviews.
One real difference: AI answers compress the market. Google shows ten organic results, ads, and a map pack. An assistant names two or three businesses and stops talking. Smaller shortlist, higher stakes per slot.
Should you invest in AI search visibility right now?
If your customers research before they buy — and in the trades, home services, and B2B they almost always do — yes, because the work is cheap relative to what it protects, and nearly all of it doubles as ordinary SEO. The businesses that should wait are the ones missing fundamentals: no real website, a half-empty Google Business Profile, no reviews. Fix those first; assistants can’t cite what doesn’t exist.
Here’s the honest picture on volume. Open GA4, go to your referral report, and look for sessions from chatgpt.com and perplexity.ai. For most small businesses that number is modest today — real, growing, but modest. So why act now?
Three reasons. First, those visitors convert unusually well, because a machine they trust just vouched for you by name. Second, the shortlist effect means early movers in a local market can occupy slots that get harder to displace as assistants accumulate consistent signals about who’s reputable. Third, the seasonal rhythm of the trades makes timing matter: if you’re an HVAC contractor, the homeowner asking Claude about furnace replacement costs in October started researching in September. Work done this summer is what gets retrieved during the season that pays your year.
There’s a decision-framework question underneath all this: is your constraint demand or visibility? If your phone rings plenty and you can’t hire techs fast enough, this can wait a quarter. If you’re watching competitors get jobs you never got a shot at, visibility is the constraint — and AI answers are a growing share of where visibility now lives.
What can you actually do this week? The checklist
Everything below is implemented on techlancer.com. This isn’t theory I’m assigning to you and skipping myself. In rough priority order:
1. Run your own visibility audit
Cost: free. Time: two hours.
Write down the ten questions your customers actually ask — “best roofer near Naperville after hail damage,” “how much does a tankless water heater cost installed,” “who builds custom software for logistics companies.” Ask each one in ChatGPT (with search on), Claude, Perplexity, and Gemini. Log who gets named and which sources get cited, in a spreadsheet, with dates.
This is your baseline. Repeat it monthly. Everything else on this list is measured against it.
2. Check crawl access
Cost: free. Time: an hour, maybe a developer’s help.
Open yourdomain.com/robots.txt and confirm GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Google-Extended aren’t blocked. This sounds obvious. It isn’t — plenty of security plugins and CDN “bot protection” settings block AI crawlers by default, and the business owner never knows. Then check that your core content is in the server-rendered HTML: view source on your homepage and search for a sentence from your main copy. If it’s not there, your site depends on JavaScript rendering that some AI crawlers handle poorly.
While you’re in there, run your key pages through PageSpeed Insights. Retrieval systems time out. Slow pages get dropped from consideration mid-answer.
3. Fix your entity everywhere it lives
Cost: free to a few hundred dollars. Time: an afternoon, then patience.
One exact business name, address, and phone number — everywhere. Your website footer, your Google Business Profile, Yelp, the BBB, your Facebook page, industry directories. Then write a real About page: who you are, where you operate, since when, what you do and don’t do. Assistants lean on About pages hard when deciding whether a business is what it claims to be. Ours says “Naperville, Illinois, building web applications since 2016” because those are checkable facts, and checkable facts are what models repeat.
4. Add structured data — and validate it
Cost: free DIY, or $1,000 to $5,000 done professionally. Time: one to three days.
JSON-LD schema is how you state facts about your business in a format machines parse without guessing: LocalBusiness or Organization, Service for each service, FAQPage where you answer questions. Write it deliberately rather than trusting whatever a plugin emits, and validate every page with Google’s Rich Results Test and the Schema.org validator. Invalid schema is worse than none — it signals sloppiness to the systems reading it.
This is the most engineering-shaped item on the list, and the one most often botched. If you hire anything out, hire this.
5. Restructure your key pages around questions
Cost: your time or $300 to $1,500 per page from a good writer. Time: ongoing.
Take your five most important pages and reshape them: headings phrased as the exact questions customers ask, with a direct two- or three-sentence answer immediately below each, then the detail. Include honest numbers — assistants love citing pages with concrete ranges, which is why a straight-answer pricing page like our website cost guide is the kind of content that gets quoted. Vague pages that say “every project is different, call us” give a model nothing to work with.
You’re reading this technique right now. Every H2 in this article that’s phrased as a question opens with a standalone answer. That’s not a stylistic tic; it’s the mechanism.
6. Add an llms.txt file
Cost: free. Time: thirty minutes.
llms.txt is a plain-text file at your site root that gives AI systems a curated map of your most important pages — loosely modeled on robots.txt. Adoption across assistants is still uneven, and I won’t pretend it’s a magic switch. But it costs half an hour, it can’t hurt, and we ship one on our own site because the downside is zero.
7. Feed the corroboration machine
Cost: mostly discipline. Time: forever, honestly.
Reviews on your Google Business Profile, replied to. Listings in the two or three directories that matter for your trade. A mention in the local paper when you sponsor the little league team. Assistants cross-reference these before naming you, the same way local SEO has always worked. If you serve homeowners, everything in our home services lead generation guide about review engines applies here unchanged.
8. Track what’s working
Cost: free to $50/month. Time: an hour to set up.
Three instruments. GA4 referral reports for sessions from AI domains. Your monthly visibility audit from step one. And your intake: add “How did you hear about us?” to every form and phone script — CallRail-style call tracking makes this systematic, but even a receptionist with a notepad beats guessing. “I asked ChatGPT” is an answer you’ll start hearing, and you want to know when.
How much does this cost, and how long until it works?
Done yourself, the technical checklist above costs a few focused days plus ongoing content time. Hired out, expect roughly $1,500 to $5,000 one-time for the technical layer — schema, crawl fixes, entity cleanup — and $500 to $2,500 a month when it’s folded into an ongoing SEO and content program. Timeline: retrieval-based citations can appear within four to eight weeks of a page being indexed; training-data mentions move on model-update timescales you don’t control.
Nobody can guarantee you’ll be cited, and you should walk away from anyone who does. Each assistant ranks sources with its own logic, which changes without notice. What’s controllable is removing every known obstacle and being the easiest business in your market for a machine to verify and quote. That’s a real edge. It just isn’t a certainty, and I’d rather tell you that now than in a quarterly report.
What mistakes waste the most money?
The expensive failures I keep seeing, in descending order of damage:
Buying “AI SEO” packages from strangers. The gold-rush vendors are here, selling $3,000/month “GEO domination” retainers that amount to a schema plugin and some AI-generated blog posts. Ask any vendor to show you their own site’s structured data and llms.txt. If they haven’t done it for themselves, keep your wallet closed.
Mass-producing AI-written content. Publishing forty thin, generated pages doesn’t make assistants cite you; it makes your whole domain look like the content farms these systems are specifically tuned to skip. Ten pages with real answers and real numbers beat a hundred pages of filler. Every time.
Blocking AI crawlers by accident. The most common self-inflicted wound. One aggressive firewall rule and every dollar spent on content is invisible to the systems you’re courting. Check quarterly — CDN and plugin updates re-break this.
Schema spam. Marking up fake reviews, stuffing keywords into JSON-LD, adding FAQPage markup to pages with no FAQs. Machines are getting better at detecting mismatch between markup and content, and trust, once lost with these systems, is expensive to rebuild.
Chasing AI visibility while the foundation rots. If your site fails Core Web Vitals and your Business Profile has four photos from 2019, llms.txt is not your problem. Sequence matters.
Should you hire help or do it yourself?
Do it yourself if you’re comfortable editing robots.txt, writing content, and following a checklist — steps one, three, six, and seven require no technical skill at all, and they’re half the value. Hire help for the schema work, for JavaScript-rendering problems, or when your time is worth more on a roof than in a text editor.
A reasonable split: DIY the audit, entity cleanup, and review engine. Bring in an engineer or an AI search optimization practice for the structured data and crawl diagnostics, ideally as part of a broader SEO audit so you’re not fixing AI visibility on a site with deeper problems. Whoever you hire, apply the vendor test from the mistakes section: make them show you their own implementation before they touch yours.
The window here favors people who move early and do the unglamorous work. Most of your competitors won’t check their robots.txt, won’t write real schema, won’t restructure a single page. That’s the opportunity. It looks like plumbing because it is plumbing — and plumbing is what the machines read.