You cannot make ChatGPT recommend your business. You can make yourself absurdly easy to cite, which is the closest thing on offer. Almost every question I get about this is three questions in a trench coat, and until you pull them apart you’ll spend money on the wrong thing. Let’s pull them apart, then I’ll show you what we changed here.
And if this is the first you’re hearing that people ask a chatbot for a web designer instead of Googling one, you’re not behind. An assistant reads a few sites live and names a few businesses. The fight is whether yours is readable.
Fair warning: our changes are recent. No chart here. I don’t have results yet, and anyone selling you results after a month is guessing louder than I am.
What do people actually mean when they say “AI found my business”?
Three separate things, each with a different lever.
One: you’re in the training data. The model absorbed something about you while it was being trained. You can’t influence that and you can’t verify it. Forty blog posts this quarter do not retroactively insert you into a model that finished training last year.
Two: an assistant retrieves you live. Somebody asks ChatGPT for a web designer in Columbus, it runs a search, reads a handful of pages, writes an answer with links. This is the one you can influence. This is the whole ballgame.
Three: you show up in Google’s AI Overviews. Those sit on Google’s ordinary search index, so the lever is ordinary SEO: pages that rank, answers to real questions, a site Google can crawl.
People sell all three as one service and call it generative engine optimization. Some of that work is real. A lot of it is technical SEO in a new hat, which is why a GEO proposal with no technical SEO should make you suspicious. We don’t sell it as a line item because it isn’t one. Crawlability, structure and schema sit inside every build under our services.
So how do you get cited by ChatGPT? Start by confirming you aren’t blocking it.
Before anything clever, check the crawlers can reach you. Cheapest check in this article. Load yoursite.com/robots.txt and read it.
The part that trips people up: each AI company runs several bots doing different jobs, listed separately. Blocking one doesn’t block the others, and blocking the wrong one costs you what you wanted.
| User agent | Who runs it | What blocking it costs you |
|---|---|---|
| GPTBot | OpenAI | Your content isn’t used to train OpenAI’s models |
| OAI-SearchBot | OpenAI | You won’t surface in ChatGPT’s search results |
| ChatGPT-User | OpenAI | ChatGPT can’t fetch your page mid-conversation |
| ClaudeBot | Anthropic | Anthropic’s crawler skips you (separate agents handle search and fetches) |
| PerplexityBot | Perplexity | You’re out of Perplexity’s cited answers |
| Google-Extended | No use in Gemini apps or Vertex AI |
One correction, because I see this constantly: Google-Extended does not control Google Search. Google’s own docs say it isn’t a way to manage how content appears in Search and isn’t a ranking signal. Blocking it doesn’t remove you from AI Overviews.
We checked ours. Nothing blocked, which is what I wanted but not what I assumed, because a security plugin or host can add rules you never wrote. Go read yours. Reading is free. Acting isn’t, because “just delete it” is how people switch off a firewall rule that was there on purpose. Send me the line instead of guessing.
Should I block AI crawlers?
Depends what your content is for. If you publish reporting or tutorials, anything where the words are the product, there’s a real argument for blocking the training crawlers. You’re giving away inventory. Legitimate call. But if you’re a plumber, a dentist, or a studio like ours, your site is a sales asset, and blocking retrieval bots to protect service-page copy is like locking your shop door so nobody reads the menu. Yes, your material may get summarized in an answer nobody clicks. Still better than not existing in it.
Who notices when this quietly breaks in six months?
Nobody. A plugin update, a host migration, a well-meaning security setting, and there’s a Disallow line nobody chose. Schema stops rendering. No error, no drop you’d spot on a Tuesday, just a slow fade out of the answers you used to be in. Which is why backups, updates and monitoring are the column that never gets cut. Drop the maintenance plan and you don’t save $300 a month, you defer it until it costs a rebuild.
Why does Bing suddenly matter again?
ChatGPT’s search has leaned on Bing’s index since launch, alongside OpenAI’s own crawling. OpenAI doesn’t publish the current mix, and independent testers keep finding other sources. Safe read as of August 2026: if Bing doesn’t have your pages, you’re gambling.
Almost nobody I talk to has opened Bing Webmaster Tools. Verify the site, submit your XML sitemap, see what’s indexed. Signing up is quick. Knowing which of the crawl errors it hands back actually cost you money is not. Free to open. Not free to act on.
Related, and this one embarrassed us: our portfolio pages had been excluded from our own sitemap. Default titles, no meta descriptions, so our best proof was invisible. Go count whether your best pages are in yours. If the number’s low, that isn’t a one-time fix, that’s how your site generates pages.
Does having one consistent description really matter?
Yes, and it’s the cheapest fix here. Write about a hundred words on what you do, who for, where, what’s different. Then use that same text on your About page, in your Organization schema, on your Google Business Profile, and in your social bios. The reasoning is entity resolution: a model reading five sources that describe you five different ways has to decide which is really you. You’re refusing to be confusing.
What schema markup should a small business actually add?
Schema is structured data in the page code, stating plainly what a human infers from the design.
Here’s what we added. The homepage got ProfessionalService schema naming our Columbus, Ohio location, service area, and a catalog of what we sell. FAQ sections got FAQPage schema. Portfolio case studies, which had none, got WebPage and CreativeWork markup.
The rule that matters: schema must match what a visitor actually sees. FAQPage schema listing questions nobody can find on the page is a lie in a format that’s trivially easy to catch.
Now, the thing I hear most. You asked ChatGPT for the JSON-LD, pasted it into a plugin field, and it looked right. Then Search Console lit up red, or it validated clean and did nothing. That isn’t you being bad at this. It’s what these tools do: produce a convincing starting point. They can’t know what’s actually rendered on your page, or whether it survives the next update. Matching markup to the real page and keeping it matched is a build decision, not a paste. Ours lives in theme templates, which is how it ships on every small business website we build.
Why put the answer in the first two sentences?
Because that’s the text most likely to get lifted. A retrieval system wants a compact chunk that answers the question it was handed. If your answer arrives in paragraph nine, after a warm-up about choosing the right partner, it’s far less likely to make the cut. So: question-framed heading, direct answer underneath, elaboration after. We restructured a pile of pages this way. Reads better for humans too, a decent sign you haven’t invented a trick.
Why does “contact us for a custom quote” give an AI nothing to quote?
Because it contains no information. A model can repeat the phrase back word for word, and the reader still learns nothing about whether they can afford you.
Compare that to our pricing page: the Launch website package is $2,400 flat, with brand kit, custom design, full build, ten pages populated, on-page SEO and training. Booking sites from $5,500. Online stores from $8,500. Maintenance at $300, $650, or from $5,000 a month, month to month, no contract.
Those are facts a machine can repeat and a human can act on. We put a calculator on that page and wrote a full piece on what a small business website actually costs for the same reason. Real numbers give somebody on their phone at 11pm something to hold onto. A pricing form doesn’t.
Is llms.txt worth adding?
We added one, and I wouldn’t spend twenty minutes on it. It’s a proposed Markdown file at your site root summarizing your business for crawlers. Not an adopted standard, and Google’s John Mueller has said flatly that no AI system currently uses llms.txt. Costs nothing, might matter later. Anybody charging real money for one is charging you for a text file. We include ours and don’t bill for it.
How do you check whether you’re actually being cited?
Ask. Genuinely, that’s the method. Open ChatGPT, Perplexity, and Google and run the questions a real customer would ask. “Web designer in Columbus Ohio.” “How much does a small business website cost.” Note whether you appear and whether the link points at the right page. Monthly, in a spreadsheet, without panicking at noise, because these answers vary between users and sessions.
Server side, your access logs show hits from those user agents, and referral traffic from chatgpt.com and perplexity.ai shows in analytics. Treat that as a floor, since everybody who read the answer and never clicked is invisible.
Now the honest part about that spreadsheet. If you’re the owner quoting jobs and running crews and doing invoicing before nine, it is not happening in month three. Not a character flaw. That’s what the hundredth item on a list does. It’s the thing you hand to somebody whose whole job is noticing it.
Questions people actually ask me about this
Can I pay to be recommended by ChatGPT?
Not for organic citations, no. Anybody promising guaranteed placement in AI answers is selling something they don’t control. What you can buy is the work underneath: a crawlable, structured site that’s honest about pricing and answers real questions. That’s a build, and that’s what we do.
Do I need a blog to get cited by ChatGPT?
No. You need pages that answer real questions clearly. A sharp services page and a useful pricing page beat thirty vague blog posts. I know that one from the wrong side, having cleaned up scaled AI content on this very blog. Getting those few pages right is the cheapest place to start with us.
How long does this take to show up?
Unknown, and I’d rather say so than invent a number. Crawlers have to re-read you, indexes update, retrieval shifts as these products change. None of it starts until the technical side is fixed, so the only clock you control is how fast you fix it.
What I’d do today, and what I’d hand off
Read your robots.txt, then your sitemap. Free, fast, and the one place a site gets locked out without anybody noticing. The rest (schema that matches the page, question-framed structure, published prices, a sitemap holding your best work, somebody watching it after the next update) is a build plus a maintenance habit. That’s not me talking you out of caring. That’s what “done” includes, because half-done looks identical to done right up until you check whether anybody’s citing you.
None of this is a hack. It’s the same instruction the web has given us for twenty years: be clear about who you are and what you charge. The machines just made being vague more expensive.
So, next step. Send me what your robots.txt says and tell me what you do for a living. I’ll tell you straight whether it’s a five-minute fix or a site that’s fighting you, no charge for the answer. Already know it’s a rebuild? Run your numbers on the pricing calculator first. I’d rather you show up knowing the budget.