Your website is not the only thing AI reads about you

In October 2026 we observed how AI assistants and search tools described companies that are not our clients. The patterns below come from those observations. We do not name the companies, and we have changed identifying details.

Your site disagrees with itself

At a US marketing services agency, the founding year in the meta descriptions of several key pages was different from the year on the About page, in the footer copyright line and on the company's directory profiles. Visitors never see a meta description on the page itself. Search engines and AI tools do read it. Some page titles had an empty space where the year should have been.

The site's structured data did not settle the question. Its Organization markup named the company and linked its social profiles, but carried no founding date and no address.

On the blog, the author of most articles was an email address rather than a person's name, both in the structured data and in the visible byline on the pages we sampled.

None of this is dramatic. Together it means a reader, human or machine, gets a different answer depending on which part of the site it reads first.

Directories keep their own version

The same agency's website gave one city for its office. Its profiles on a professional network, a review platform and a B2B directory gave an address in a different city. The phone number was the same everywhere, and the website did not explain the relationship between the two locations.

In late October 2026 we asked Perplexity, logged out and once, to describe the company. The answer took the founding year and the city from the website, and then told the reader that directory listings still showed the other city. The contradiction reached the person asking, inside the answer. The assistant did not decide which version was true. It reported both.

A page you want read is closed to the readers

The agency's robots.txt file contained a short Disallow rule. Google's robots.txt documentation explains that a rule matches every path that begins with the given text: a rule for /fish also matches /fishheads. Here the rule also covered the case studies section, a page the site linked from its main menu and marked for indexing in its own meta tags.

When we asked assistants to open that page, most of them could not. The ones that could reach the site at all opened its other pages without trouble. The page holding the company's own proof of work was the one page those assistants could not read.

This is the quietest version of the pattern. Nothing on the site looks broken to a visitor. The site simply sends two opposite instructions, and the stricter one wins.

Forgotten copies keep speaking

Test, staging and old campaign subdomains are easy to forget because nobody inside the company visits them. Search engines keep them as long as they stay reachable. Blocking such a host in robots.txt does not take it out of search. Google's own guidance says that if robots.txt blocks a page, its crawler never sees the noindex rule, and the page can still appear in results. To keep a page out, Google lists removing or updating the content, password protection, or a noindex rule the crawler can read. Its Removals tool is temporary: requests last about six months.

Five checks you can run this week

1. Look for forgotten hosts. Search Google for site:yourdomain.com and look through the results for subdomains you do not recognise: test, staging, dev, old campaign or product names. Google notes that the site: operator does not necessarily return every indexed URL, so treat this as a first look and confirm individual addresses with the URL Inspection tool in Search Console.

2. Compare robots.txt with your meta robots tags. Open yourdomain.com/robots.txt. For every Disallow line, ask which of your important pages begin with that text, including case studies, pricing, team and contact pages. Then open those pages and check their meta robots tag. A page marked index in its own tags but disallowed in robots.txt is sending two opposite instructions.

3. Put your directory listings next to your contact page. Open your company page on professional networks, review platforms, B2B directories and map services. Compare the company name, address and phone number with your website, word for word. Where they differ, decide which version is correct and update the others, or explain on your own site why there are two locations.

4. Read your author fields. Open a blog post, view the page source and search for "author". The author should be a named person or the company itself, not an email address or a placeholder, and it should match the byline a visitor sees.

5. Check your founding year everywhere. About page, footer, page titles, meta descriptions, structured data and every directory profile. One year, written the same way, in every place.

Method note

Observation period: October 2026. We used public AI assistants and search tools, logged out where the service allowed it. Assistant answers change between runs, between sessions and over time, so a single answer is an observation, not a measurement of how an assistant always behaves. The companies we observed are not TKM Studio clients and did not commission or review this article. Identifying details have been changed.

Sources

• ·Google Search Central, How Google Interprets the robots.txt Specification: https://developers.google.com/search/docs/crawling-indexing/robots/robots_txt
• ·Google Search Central, Block Search Indexing with noindex: https://developers.google.com/search/docs/crawling-indexing/block-indexing
• ·Google Search Central, Remove a page hosted on your site from Google: https://developers.google.com/search/docs/crawling-indexing/remove-information
• ·Google Search Central, How To Use the Site Search Operator: https://developers.google.com/search/docs/monitor-debug/search-operators/all-search-site