Why AI assistants get your business wrong
If an assistant is quoting a price you stopped charging in 2023, nobody has lied to it. It read what was there.
More of your customers are asking an assistant before they ask you. They type a question into ChatGPT, or ask Gemini on their phone, or get an AI summary at the top of a Google page they never scroll past. Sometimes the answer about your business is wrong.
The instinct is that the AI made it up. Occasionally it did. Far more often it read something you published, believed it, and repeated it faithfully. The fix is usually not clever: it is giving the machine something correct to read.
What an assistant is actually doing
When an assistant answers a question about a named business it is not recalling a fact it was taught. It is fetching pages, reading them, and summarising. That means three things decide what it says: what it can reach, what it can parse, and whether the sources agree with each other.
Each of those fails in a different way, and each has a different fix.
One: it cannot reach you
Your site has a file called robots.txt that tells automated visitors what they may read. Most sites have never been looked at. A good many quietly block the AI crawlers, either because a plugin did it or because a well-meaning setting called something like "block AI training" was switched on.
That setting does more than you think. Blocking GPTBot means ChatGPT cannot read you. Blocking Google-Extended means Gemini and Google’s AI summaries cannot. The assistant then answers from whatever it can find elsewhere: a directory listing, an old news item, a competitor’s comparison page.
There is a real decision underneath this. Blocking those crawlers does keep your writing out of training data. It also keeps you out of the answers. For most small businesses the second cost is far larger than the first, but it is your call to make deliberately rather than by default.
Two: it cannot parse you
Being readable is not the same as being reachable. A page can load perfectly for a person and still give a machine almost nothing to work with.
The common causes are dull and fixable:
- Your prices, hours or services exist only inside an image or a PDF, where there is no text to read.
- The page is built so the content only appears after scripts run, and what arrives first is an empty shell.
- Nothing on the page is marked up as structured data, so a price is just characters near a dollar sign rather than a price.
- The answer to the obvious question is buried in the middle of a paragraph about something else.
Structured data is the one worth understanding. It is a small block of machine-readable description sitting alongside your page, saying this is the organisation, this is the service, this is the price, this is the area it serves. An assistant does not have to infer any of it. On our own site, the half-day programme is priced at $2,180 plus GST in the copy and separately marked up as a number with a currency, so there is nothing to guess at.
Three: your sources disagree
This is the one that catches established businesses hardest, and it is almost always the real cause when an assistant quotes something out of date.
Your business exists in more places than your website. A Google Business Profile. LinkedIn and Facebook. Industry directories. An old site on a domain you forgot to redirect. A listing somebody created for you years ago. Each one carries a version of your name, address, phone number and what you do.
When those disagree, something has to choose. A search engine treats the inconsistency as a reason to trust all of it less. An assistant picks one, and it will not necessarily pick the one you would have.
What actually fixes it
Four things, roughly in the order they pay off.
Let the crawlers in, on purpose
Name them in robots.txt and allow them explicitly rather than relying on a wildcard. Some check for a rule that names them before indexing at all. This is twenty minutes of work and it is the difference between being quotable and being absent.
Answer the question in the open
Put the question somewhere a machine can see it is a question, and the answer directly beneath, in plain words. A page that says "How much does it cost?" followed by a number gets quoted. A page that says "Investment" followed by three paragraphs of positioning does not.
Mark up the facts
Your organisation, your services, your prices where they are public, your frequently asked questions. This is the layer that turns characters into facts, and it is invisible to your visitors, so there is no design cost to weigh against it.
Make every listing agree
One name, one phone number, one address, everywhere. Link your social accounts from your own site, and list them in your structured data, so the claim that they are yours has something supporting it.
How long it takes to show up
Slower than you would like. Crawlers come back on their own schedule, and assistants vary from days to a couple of months. There is no button that forces a re-read.
What you can do is make sure that when it happens, what they find is current, consistent and legible. That is the whole job. It is not glamorous and it does not need to be clever — it needs to be true and readable, which is a lower bar than most businesses are currently clearing.
This is most of what we do when a site is rebuilt.
How we rebuild a site to be read