Why AI doesn't cite your website: it's often about the date
A page that hasn't been updated recently, or whose date a machine cannot read, loses its spot in AI answers and in Google. A stale or unreadable date cuts the odds of reaching the top 3 by 56%, and 83% of AI citations on commercial queries come from pages updated within the past year.
You have a good website, your services are well explained and your clients are happy. Yet when a buyer asks ChatGPT who to hire in your field, another company's name comes out.
The cause is often simpler than people think: your page hasn't been touched in three years.
This isn't a theory. Two independent analyses published this summer reach the same verdict and it's the kind of problem you can fix in a morning.
What counts as "stale" content for an AI?
A stale page is a page that hasn't been updated recently or whose date is impossible for a machine to read.
The second case surprises a lot of people. All three of these situations count as "stale":
- No date displayed anywhere on the page.
- A date written as vague text: "updated last fall".
- A date visible to a human but absent from the page's code, therefore invisible to the robots reading it.
In that last case, your page may have been rewritten last month: if the information isn't in the code, the AI treats it as old.
What does the data actually say?
Kevin Indig, a well-known SEO analyst, published a study in August 2026 covering 60,000 queries and nearly 60,000 Google results pages.
Well-ranked pages are markedly fresher than the rest:
| Top 3 pages | Pages in positions 8 to 10 | |
|---|---|---|
| Date from the past two years | 66.6% | 57.3% |
| Old or unreadable date | 8.2% | 15.1% |
The number to remember: once other factors are controlled for, an old or unreadable date cuts the odds of reaching the top 3 by 56%.
Separately, the AirOps team analyzed thousands of AI-generated answers. Their finding points the same way: 83% of AI citations on commercial queries come from pages updated within the past twelve months, and a stale page is three times more likely to lose the citations it already had.
Two different methods, two different teams, same conclusion.
What makes the result striking is what did not survive. Indig tested more than fifteen factors across 3,000 pages: text length, image count, author bios, technical markup, number of subheadings. Once the other variables were controlled for, almost none held up. Freshness is the only signal left standing.
The lesson simplifies the work: you don't need to rebuild your website, you need to keep your important pages up to date.
Does changing just the date work?
No. And this matters, because it's many people's first reflex.
Changing the date without touching the content is a trick search engines and AIs detect. They compare the declared date against what actually changed. A site that keeps pushing its dates forward without substance loses credibility, which is exactly the opposite of the intended effect.
The right sequence is always the same: refresh the content, then update the date.
Refreshing doesn't mean rewriting everything. It means correcting prices that changed, removing a service you no longer offer, adding the two cities you've been serving since last year, replacing a 2023 figure with this year's.
How to tell if your pages are stale?
Three quick checks, no paid tools.
1. Open your service pages.Do you see a date displayed? If not, that's already a problem.
2. Look for information that has aged. A price, a timeline, a list of cities served, a discontinued service. If you find something you would no longer say on the phone today, the page is stale.
3. Ask ChatGPT or Perplexity to describe your business.If the answer contains outdated information, you're seeing live what the AI retained from your old page.
That third test is the most revealing, and it's often the one that makes business owners react.
What this changes for a Quebec SMB
Here is my own example. I run no advertising for my business. My website is shared in two places: my LinkedIn profile and the link in my Instagram bio. That is it.
For the past month, I have been getting one to two inquiries a week, and it is almost always the same reason: the person asked ChatGPT a question and my name came up. They never compared me inside a list. They asked, they got two or three names, they wrote.
My reports, on the other hand, know nothing about it: those visits show up as "Direct". I explain why in your clients come from ChatGPT, Google Analytics says "Direct".
That is exactly the shift under way. Before, Google showed an article like "the 5 best contractors in the area" and the client clicked through to read. Today, the AI names a company directly, and you have to be that name.
That means two things. First, being in the top 10 is no longer enough. Second, the page the AI consults to talk about you must reflect your business as it is now.
A website you put online and then forget was a fine website in 2019. In 2026, it's a website quietly disappearing from the answers.
Frequently asked questions
How often should I update my website?
Aim to review your commercial pages every six to twelve months. The data shows 83% of AI citations on commercial queries come from pages updated within the past year, so twelve months is the limit not to exceed on those pages.
Do I need to rewrite my pages entirely?
No. Correcting prices, services offered, service areas and dated figures is enough in most cases.
My website is recent. Does this concern me?
Yes, in a year. A new site starts with the freshness advantage, but that advantage wears off on its own if nobody touches it.
Does this replace classic SEO?
No, it adds to it. Freshness helps both your Google rankings and your AI citations. It's one of the rare projects that serves both.
Want to know where you stand?
I run audits that check exactly this: what AI sees on your website, what it says about you and which pages make it hesitate. Based in Greater Montreal, working with businesses across Quebec.
Request an auditSources: Kevin Indig, Growth Memo, August 2026 (60,000 queries, 3,000 pages crawled); AirOps, analysis of AI-generated answers, August 2026.