Last updated: July 2026
You ask ChatGPT to recommend businesses in your own industry and city — and it names your competitors. Or worse, it describes your company using details that stopped being true years ago. Meanwhile a growing share of customers start exactly there: asking an assistant instead of scrolling a results page.
The good news is that AI invisibility almost always has a mundane cause. Below are the six we see most, roughly in the order you should check them. Each one comes with a way to diagnose it and a fix.
1. Your site is blocking the bots that answer questions
The single most common cause, and the easiest to have caused by accident. Somewhere along the way — a plugin, an agency, a bot-protection service, a well-intentioned robots.txt edit — your site started blocking AI user agents. The subtlety is that there are two kinds of AI bot, and most blocking advice ignores the difference:
- Training crawlers like GPTBot and ClaudeBot collect data to train future models. Blocking them is a legitimate choice with little downside for visibility today.
- Search and on-demand fetchers like OAI-SearchBot, ChatGPT-User, and Claude-User visit your site at the moment a user asks about you. Block these and you are invisible in live AI answers — the assistant literally cannot read your site.
Many sites block everything with "AI" in the name and unknowingly opt out of being recommended.
Diagnose: run your domain through the free AI crawler robots.txt checker — it shows per bot whether your robots.txt allows or blocks it. Also check any firewall or CDN bot-protection settings, which can block bots without touching robots.txt.
Fix: allow the search/fetcher bots; decide separately and deliberately about the training crawlers. Our AI bots directory lists each one with copy-paste robots.txt rules.
2. Your content doesn't exist without JavaScript
Many AI fetchers grab your page's raw HTML and read what's there. If your site is built so that the actual text only appears after JavaScript runs — common with certain site builders, headless setups, and "animation-first" templates — a fetcher may see a skeleton with no content at all.
Diagnose: open your homepage, hit View Page Source (not inspect), and search for a sentence you know is on the page. If your services, address, and descriptions aren't in the source, a good share of bots can't see them either.
Fix: server-side rendering or static generation if your platform supports it. Mainstream platforms like WordPress, Webflow, Squarespace, and Shopify render text into HTML by default — if you're on one of those, this usually isn't your problem.
3. Your site never actually says what you do
Assistants can only repeat what you state. A surprising number of business sites never say — in plain text — what the business does, where it operates, and for whom. The information lives in a logo, a hero image, a PDF price list, or the owner's head.
Diagnose: read your homepage pretending you know nothing about the company. Can you answer: what do they sell, where, for whom, at roughly what price? If a human can't, an AI can't.
Fix: write it down, boringly and explicitly. One page per service. Location in text. Prices or ranges. This overlaps entirely with step 2 of our guide on getting recommended by ChatGPT — the same fixes serve both.
4. You don't exist outside your own website
Before recommending a business, assistants look for corroboration: reviews, directory listings, maps profiles, mentions on other sites. If the only place on the internet that says you exist is your own domain, you look — to a cautious machine — indistinguishable from a website someone made yesterday.
Diagnose: search your business name plus your city in a private browser window. If nothing but your own site comes up, that's the gap.
Fix: complete Google Business profile, the directories relevant to your industry and region, and a steady trickle of reviews. Unofficial but real: a couple of mentions in local press or industry blogs go a long way. This is the slowest fix on the list and the one with the most compounding value.
5. Your information is stale or self-contradictory
AI assistants fetch and repeat. If your site says one set of opening hours and your Google profile says another, the assistant picks one — possibly the wrong one — and tells the customer. Old prices, closed locations, and abandoned services all get recited with full confidence.
Diagnose: ask ChatGPT (with web search enabled) and Claude: "What do you know about [business name] in [city]?" Read the answer carefully. Everything wrong in it is something wrong — or ambiguous — in your public footprint.
Fix: a quarterly check of site, Google profile, and directories. Consistency matters more than polish.
6. The assistant has no fast way to understand your site
This one is about friction rather than blockage. A fetcher answering a live question has seconds and a limited budget of pages. On a site with dozens of pages, it has to guess which ones define your business — and it can guess wrong, building an answer from your blog archive instead of your services page.
An llms.txt file removes the guesswork: one plain-text file at yourdomain.com/llms.txt summarizing what your business is, with a curated list of the pages that matter.
We're deliberately putting this last. As we lay out honestly in Does llms.txt actually work?, it is not a magic visibility switch, most training crawlers ignore it, and it will not fix problems 1 through 5. It's the finishing touch for a site that's already reachable and clear: cheap to add, read by exactly the kind of on-demand fetchers that answer customer questions, and useful to any agent that lands on your domain.
Fix: generate one here — paste your URL and you'll have a file in minutes. Already have one? Validate it.
Work through the list in order
The order isn't arbitrary — it's a dependency chain. An llms.txt file can't help if bots are blocked (1). Reachability doesn't help if there's nothing readable (2, 3). Clarity on your own site rings hollow if nothing external backs it up (4), and everything unravels if the information is stale (5).
Most businesses that feel "invisible to AI" have a problem in the first three, and those are all fixable in a week.
Frequently Asked Questions
How do I see what AI bots are doing on my site right now?
Your server or CDN access logs list every visit by user agent — look for names like ChatGPT-User, OAI-SearchBot, and Claude-User. Our AI bots directory has the exact user-agent strings to search for.
I fixed everything — how fast will AI answers change?
Answers built on live web search can improve immediately, since the assistant re-reads your site each time. The model's built-in memory of your business updates only when new models are trained, so residual outdated "knowledge" can linger — one more reason your live site needs to be unambiguous.
Is this different from normal SEO?
Mostly no, and that's encouraging: reachability, clear content, consistent listings, and reviews are classic SEO work that now pays a second dividend in AI answers. The genuinely AI-specific pieces are the bot-level robots.txt decisions (1) and llms.txt (6).
Should I block AI training crawlers or not?
That's a business decision, not a technical one — blocking GPTBot or ClaudeBot keeps your content out of future training runs without affecting live search answers. Just don't block the fetchers by accident while you're at it. We cover who does what in the AI bots directory and in llms.txt vs robots.txt.