7 signs that your website is invisible to ChatGPT

Your website may look perfect in your own browser but still be invisible to ChatGPT, because AI crawlers read the raw code without JavaScript and without images. Here are the seven signs you can check yourself in two minutes each — robots.txt, an empty page without JavaScript, no prices, no structured data, the name spelled differently, a title lacking ‘what’ and ‘where’, and no answers to customers’ questions — along with what each issue costs you and how to fix it.

Your website may look perfect in your own browser and still be invisible to ChatGPT. The difference is that AI crawlers read the raw code — without JavaScript, without images and without making any assumptions. Here are the seven signs you can check yourself in two minutes each, what each sign is costing you, and how to fix it.

Contents:

• Sign 1: your robots.txt says ‘no’ to GPTBot
• Sign 2: your homepage is an empty shell without JavaScript
• Sign 3: Your page shows no prices
• Sign 4: Your page has no structured data
• Sign 5: your company name is spelled differently in various places
• Sign 6: Your title doesn’t say what you do — or where
• Sign 7: Your website doesn’t answer customers’ questions
• All 7 signs at a glance (table)

Why can’t ChatGPT see a website that looks fine?

Because ChatGPT doesn’t look at your website — it reads it. When someone asks ChatGPT, Perplexity or Google AI Overviews about a plumber in Aarhus or an accountant for a small online shop, the search engine sends a crawler (GPTBot, PerplexityBot, Google-Extended) to retrieve the raw HTML. It doesn’t run JavaScript, doesn’t read text in images and doesn’t guess what a company is called. Anything that isn’t text in the code doesn’t exist for it.

This is not a niche problem. Our own scan of 465 search-visible Danish business websites on 21 July 2026 showed that 89 per cent lacked a llms.txt, that 5 per cent (23 sites) blocked GPTBot in their robots.txt, and that 5 per cent (22 sites) had virtually no readable text when JavaScript was not running (full analysis and methodology). None of the site owners could see this in their own browser — because the browser runs JavaScript, whereas the crawler does not.

What does it cost you to be invisible?

The answer lies in where the customer stops. In July 2025, the Pew Research Centre analysed 68,879 real searches and found that only around 8 per cent click through to a website when an AI summary appears at the top — compared with 15 per cent without one — and that approximately 1 per cent click on a cited source (Pew’s survey). Being mentioned in the answer is the new form of visibility. The click comes afterwards, if it comes at all.

And it is your own site that the answer is based on. In our own survey of 576 AI answers about the same 465 companies, 61 per cent of the 4,034 cited sources were the company’s own website (the survey and its limitations). If your site cannot be read, the most important source of information about you is lost — and the search engine will find another one.

Sign 1: your robots.txt says no to GPTBot

A single line in that small file, which nobody looks at, can shut ChatGPT out of your entire site. Most sites inherit this line from a template or a ‘block AI’ plugin, without anyone having decided to do so. In our own scan, this applied to 5 per cent of the 465 Danish sites, and accountants were affected three times as often as plumbers (figures per sector and per crawler).

Here’s how to check (2 minutes): type your domain into the browser followed by /robots.txt — for example, ditfirma.dk/robots.txt. Look for the two lines User-agent: GPTBot and Disallow: / one after the other. If they’re there, ChatGPT has been blocked. Repeat this for PerplexityBot, ClaudeBot and Google-Extended. If you get an error page because the file doesn’t exist at all, everything is accessible — that’s fine.

What it costs you: everything. The block applies to the entire site, not just one page, and it overrides all other optimisations: a page that the crawler is not allowed to fetch cannot be read, understood or cited, no matter how good it is.

How to fix it: delete the two lines – either manually or via your SEO plugin (in Yoast and Rank Math, robots.txt is found under ‘Tools’). If the block is a deliberate choice because you do not want your text used for training, it is set correctly — but the price you pay is being mentioned in AI responses. Geoa measures access for each AI crawler in the free trial and prioritises this fix in the queue.

Sign 2: your homepage is an empty shell without JavaScript

Many modern site builders and pages built in React, Vue or Angular without server-side rendering display a smart-looking homepage to humans — and an empty shell to the crawler, because the text is only fetched by JavaScript in the browser. AI crawlers do not launch a browser. In our own scan, 22 of the 465 sites (5 per cent) were effectively empty without JavaScript (this is what an invisible page looks like in the code).

Here’s how to check (2 minutes): open your homepage, right-click and select ‘View source’ (Ctrl+U on Windows, Cmd+Alt+U on Mac). Press Ctrl+F and search for a complete sentence from your homepage — for example, your first headline. If you can’t find it in the code, it won’t be there for ChatGPT either. If you want to see exactly what the crawler sees, the free AI readability check shows your page through GPTBot’s eyes.

What this costs you: everything hidden behind JavaScript — services, prices, address, reviews. The page exists for Google, which runs JavaScript with a delay and is invisible to the AI engines at the same time — which is why nobody notices it.

How to fix it: it rarely requires a new website. The three common approaches are server-side rendering (SSR), pre-rendering the most important pages, or placing the core content — who you are, what you do, where and what it costs — directly in the HTML. Ask your agency one question: “What does a crawler see on our homepage without JavaScript?” If they can’t answer, you have your answer.

Sign 3: your site shows no prices

The question of price is one of the most common that customers ask an AI: “How much does an accountant cost for a small online shop?”, “How much does it cost to have a mixer tap replaced?”. If there’s no price listed on your site, the search engine will fetch the figure from a price comparison site or a competitor — and then it’s those figures that get quoted. In our own measurements, companies that are quoted mention specific prices more often than those that aren’t.

Here’s how to check (2 minutes): open your main services page and type Ctrl+F. Search for “kr”. No results — or only “contact us for a quote” — is a red flag. Then search for the unit customers ask about, e.g. “per hour” or “per m²”: the page may list prices but still lack the figure for exactly what is being asked about.

What this costs you: you’re missing out on the enquiry that’s closest to a sale. The AI still responds — just with someone else’s figures.

How to fix it: list indicative prices, starting prices or three price examples, stating the relevant conditions (materials, VAT, substrate). You don’t need a price list for everything — you just need one figure that a machine can quote for the three things customers ask about most often. Geoa’s AI employee drafts the text based on your own information, and you approve it before it goes live on the site.

Sign 4: your page has no structured data

Structured data (schema.org, written as JSON-LD in the page’s code) is the part of the page written for machines: name, address, telephone number, opening hours and sector, in a format ChatGPT and Google do not have to guess. Without them, the search engine has to deduce who you are from free-form text — and if it guesses wrong, the incorrect information will be repeated. The same problem arises when the address and telephone number are only shown in an image: a crawler cannot read text in images.

Here’s how to check (2 minutes): view the source code (Ctrl+U) and search for application/ld+json. No results means it’s fine. If you find a block, read it: does it say Organization or LocalBusiness with your actual name and address — or just an empty WebSite that the theme has inserted? Finally, check that the address and telephone number appear as text that you can select with the mouse.

What it costs you: incorrect or missing basic information in the response — wrong town, wrong name, no opening hours — and a weaker link between your site and the other places where you’re listed: CVR, LinkedIn, Trustpilot, Krak.

How to fix it: a single JSON-LD block with Organisation or LocalBusiness: name, legal name, address, telephone number, email, opening hours and ‘sameAs’ links to the CVR listing and your profiles. Most CMSs have a plugin for this; Geoa’s AI employee generates the block based on what it knows about you and adds it to the site once you’ve approved it.

Sign 5: your company name is spelled differently in various places

“Hansen VVS ApS” in CVR, “Hansen-VVS” in the footer, “hansenvvs” on Facebook and “Hansen VVS & Blik” on Google Business Profile. To a human, it’s the same company. To a language model, it’s four candidates, each getting a share of the mention — and eventually it deduces the name from your domain instead. Geoa measures precisely that: if the AI spells your name differently to how you do, it doesn’t have the name from a source it trusts.

Here’s how to check (2 minutes): enter your company name as it appears in the footer of your website. Then look it up in four places:

• The CVR Register (datacvr.virk.dk) — the legal name
• your Google Business Profile — the name shown on the card
• your LinkedIn page
• your Trustpilot or Krak profile

If the spelling isn’t identical in all four places — including spaces, hyphens and company type — there’s a problem.

What this costs you: mentions that aren’t credited to you. A language model can only attribute a mention to you if it can resolve it to you.

How to fix it: choose one canonical spelling and use it in the title, footer, ‘About’ page and all profiles. Declare the variants that already exist, rather than hiding them: list them as `alternateName` in the Organisation schema, add your CVR number as an identifier and link to the register using `sameAs`. Then the system can resolve all four spellings to you.

Sign 6: your title doesn’t say what you do — or where

The browser tab displays the page’s title tag; that and your main heading are the two lines a search engine reads first. “Home”, “Welcome” or just the company name say nothing about the fact that you’re an electrician in Kolding. The town in the title tag is one of the signals we measure: the majority of local businesses that are listed have it.

Here’s how to check (2 minutes): open your homepage and read the text in the browser tab. Then read the main heading on the page. Can a stranger tell, based on those two lines alone, what you do and where? Also check the meta description: view the source code and search for name="description". If it’s missing or empty, that’s another sign.

What this costs you: you won’t be matched in ‘people nearby’ searches: the search engine can read the entire page and still fail to place you in the town the customer is asking about.

How to fix it: Title: what you do in [city] — company name. Headline: the same information in one sentence. Meta description: two sentences covering your service, the town and who you help. This is the cheapest fix on the list; you’ll find it in any CMS under “SEO” or “page settings”.

Sign 7: your page doesn’t answer customers’ questions

Customers don’t ask the AI for your company name. They ask “how much does … cost?”, “how do I choose …?”, “can I … myself?” and “who is near …?”. If the answers aren’t on your site as text, the search engine has nothing to cite you for — and so it cites whoever wrote the answer. The Princeton study behind the concept of GEO (Aggarwal et al., KDD ’24) measured up to 40 per cent higher visibility in generative search engines for pages that include sources, statistics and citations (arXiv:2311.09735) — provided there is actually something to cite.

Here’s how to check (2 minutes): write down the three questions you’re most often asked on the phone or by email — in the customer’s own words, not yours. Search for the keyword from each question on your site using Ctrl+F. If there isn’t a section that answers it, that’s a red flag. If you’re short of questions to start with, you’ll find them in ‘10 questions your customers ask’ (ChatGPT).

What it costs you: this is where ongoing visibility is lost — once per question, every time it’s asked.

How to fix it: one FAQ section per service page, with three to five questions in the customer’s own words and a direct answer of 40–60 words beneath each one — including figures, assumptions and location. This is the format that a machine can most easily quote. Geoa’s Content Studio writes these kinds of answers based on your own information — five articles a month are included in the Starter plan.

How do you check all seven signs in a quarter of an hour?

Tackle them in order. The first two determine whether the rest can even be read; the next five are plain text that you can edit in your CMS without having to ring anyone.

What do you do when you’ve found a character?

Correct them in the order they appear — the foundation before the text, then check again before moving on to the next one:

1. Elements 1 and 2 first. These usually require the person building the site — send them the checklist above so they know exactly what needs changing.
2. Fix elements 3 to 7 yourself. These are text and settings, not code; most site owners can sort them out in an afternoon using their own CMS.
3. Check again after each correction. Otherwise, you won’t know what worked — and you won’t be able to see the correction in your own browser.

At the same time, be honest about what the indicators can do. They determine whether you can be read and cited — not whether ChatGPT recommends you over your neighbour. In our own analysis of 576 AI responses, the technique was a strong predictor of whether a page was cited, but said nothing about whether the company was recommended; that is determined by reputation, reviews and media coverage (the difference between being cited and being recommended). The fundamentals first — that’s the part you can fix yourself today.

If you don’t want to check it manually, the free test at geoa.app analyses your site across 15 technical dimensions — including AI crawler access, readable text without JavaScript, structured data, title and meta description — in 60 seconds without needing to log in. Geoa’s AI employee will then fix whatever can be fixed on your site, and you approve each correction via chat or email. The corrections are included in all paid plans from €39/month, with a 14-day free trial with no credit card required. You can find out how the score works in ‘How to improve your GEO Score’.

What are other owners asking about?

Can I see the problem in my own browser?

No, and that’s the whole point. Your browser runs JavaScript and displays images; the crawler does not. A page can appear complete to you and empty to GPTBot at the same time. That’s why you should check the source code or use a tool that fetches the page without JavaScript — never by simply looking at it.

Is it enough to unblock GPTBot?

No. Access is a prerequisite, not the solution. Once the crawler is allowed in, there must be readable text, a unique name, prices and answers to retrieve — otherwise it has access to nothing. Fix character 1 first, because nothing else will work until that’s done, and then work your way down the list.

Do I need a llms.txt?

It’s a minor signal, and it doesn’t do any harm: 89 per cent of the 465 sites in our own scan were missing it. But it doesn’t replace characters 1 and 2 — a llms.txt won’t help a crawler that’s been locked out or encounters a blank page. Create it once the first two characters have been corrected; it takes half an hour using the guide here.

How quickly does ChatGPT detect a correction?

It depends on the engine. Perplexity and ChatGPT, with search enabled, fetch the page when the query is made, so a correction can be applied as soon as the page is crawled again. Responses based on the model’s training data only change at the next training session, and this isn’t controlled by anyone externally. That’s why we never promise a specific outcome from someone else’s AI — we make the page readable and quotable; the rest is up to the search engine.

Start with the first sign today. Check your website for free at geoa.app/gratis-geo-test — 60 seconds, no login required — and see which of the seven signs apply to you, and what needs to be corrected first.