A company retires a product, changes its pricing, and moves on. The web page announcing the old version doesn’t get the memo. It just sits there — still live, still indexed, still waiting for an AI chatbot to read it, quote it, and repeat it the next time someone asks about that brand.
Most AI-visibility advice skips straight past this. Everyone wants to know how to get an AI model to mention their brand more. Almost nobody asks the more basic question first: what is the model already reading, and how much of it is years out of date?
Why AI tools keep quoting pages your brand forgot about
ChatGPT, Perplexity, and Gemini answer questions using two kinds of material: what they learned during training, and what they can pull from the live web when they answer. A page doesn’t need to be new, popular, or even linked from the homepage anymore to show up in either one. It just needs to have been readable once, and useful enough to remember.
So changing a “last updated” date doesn’t undo what a model already learned from a page. And deleting the page doesn’t erase what the model picked up before it disappeared.
There’s a real example of exactly this. The Content Marketing Institute, a publisher that covers content strategy for marketers, reported on it. Sitecore, a company that makes website and digital-experience software, was doing a routine content cleanup. The team found an old page already marked for removal. That page turned out to be one AI tools cited often when answering questions about the brand. They didn’t delete it. They updated it, kept it live, and fixed what needed fixing. Sitecore’s CMO (Chief Marketing Officer), Michelle Boockoff-Bajdek, put the lesson in one line: “Age alone does not tell you what to retire.”
That’s the whole problem, really. A page’s age or traffic tells you how it performs with human visitors today. It tells you nothing about whether an AI model has already decided that page is trustworthy and is still quoting it.
Why this matters more than most brands assume
Easy to file this under “small, technical worry” — except the research says otherwise, at least for the US market these studies cover.
Forrester — a research and advisory firm that surveys business buyers and tracks how they make purchase decisions — wrote in January 2026, based on its own 2025 Buyers’ Journey Survey, that 94% of business buyers now use AI somewhere in their buying process. That’s not a niche behavior anymore. It’s close to standard.
Skyword, a content-marketing company, commissioned a separate survey in April 2026. Dynata, a firm that fields online surveys for researchers, ran it among 1,000 US adults. It found that 54% of people look to outside sources to compare information when an AI answer seems to conflict with what a brand says about itself. When the AI’s version of a brand doesn’t match the brand’s own story, people don’t shrug and move on. They go check elsewhere. An outdated page feeding an AI model the wrong story actively sends curious buyers looking for a second opinion.
A follow-up Skyword/Dynata survey from October 2026 adds one more wrinkle: 45% of US adults said they’ve had trouble choosing between companies because AI-generated answers made the companies sound too similar. Worth being precise here — these are US consumer and B2B buyer numbers, not Indian or APAC figures, and they’re used to show why this problem deserves attention, not as a claim about how Indian buyers behave. But the mechanism isn’t US-specific. If an AI model is quoting old, generic, or outdated pages, a brand’s actual differences get flattened into sameness in any market.
The audit most AI-visibility pitches skip
Most pitches for “AI visibility” or Generative Engine Optimization (GEO) work start with how to get an AI model to mention a brand more. That’s the wrong starting point unless someone has already checked what the model is saying, and why.
A simple, honest audit looks like this:
- Ask the real questions buyers would ask. Put the actual buyer-style prompts about the brand to ChatGPT, Perplexity, and Gemini, more than once, since answers shift between runs. Save what each tool says, and save any sources it names or links to.
- List every page touching products, pricing, or positioning, including the ones marked for deletion. This includes pages nobody has opened in years, old pricing pages, and anything describing a plan or feature that no longer exists.
- Cross-check the two lists. If a page from step 2 shows up as a cited source in step 1 or its old wording is clearly shaping how the AI tool talks about the brand that page isn’t dead. The model still trusts it, whatever the traffic report says.
An Indian SaaS brand’s content archive, as an example
Picture a mid-sized Indian SaaS (Software as a Service) company — a project-management tool, nothing real, purely illustrative. In 2022, it published a blog post walking through its pricing plans, including a “Starter” tier at ₹999 a month. A year ago, the company retired that plan and raised prices. Nobody deleted the 2022 post. It still ranks decently and still pulls a trickle of organic traffic. On paper, it looks like a minor, low-priority page — the kind nobody circles back to during a cleanup, because nothing about its numbers screams “fix me.”
That’s exactly the trap. The post was well-written, well-linked, and heavily indexed in its time — exactly what makes a page easy for an AI model to find and trust. Apply the three-step audit above to this brand, and there’s a real chance the answer to “what does this tool cost” leans on that 2022 post, quoting a plan and a price that stopped existing twelve months ago. A traffic dashboard would never flag this page as urgent. Checking what ChatGPT or Perplexity actually says would.
This is the Sitecore situation almost exactly. The instinct during a cleanup is to delete the old post. The current pricing page already exists, so the 2022 one looks redundant, even embarrassing. But a flat delete throws away whatever authority and citation history that page had already earned, and closes off any chance to fix what it’s actually telling people. The better move is almost always to go back into the old post directly: correct the plan and price, add one clear line noting what changed, and let the page keep doing its job accurately, this time.
Four honest options for every outdated page (and how to choose)
Not every outdated page needs the same fix. Four honest options, in practice:
Rewrite in place. The right call when a page still ranks or still gets cited, and the only real problem is stale facts — the Indian SaaS pricing post above, Sitecore’s own example, both fit here. Keep the page, fix what’s wrong, move on.
Consolidate. For when two or three old pages say overlapping, slightly contradictory things about the same topic. One brand might have a 2021 FAQ, a 2023 pricing post, and a current pricing page all live at once, each telling a slightly different version of the same story. Merge them into one accurate page, and redirect the rest into it.
Retire with a redirect. For pages that are genuinely obsolete, where nothing current needs to point back to them. A redirect — never a flat deletion, carries forward whatever ranking or citation value the old page built, instead of letting it vanish.
Leave as-is, clearly dated. The rare case, but a real one. A page honestly framed as a point-in-time update — something like “here’s what changed in our product in 2021” — isn’t misleading just because it’s old, as long as nothing current contradicts it.
Google’s own guidance backs up the instinct to slow down before deleting anything. Its documentation on creating helpful, reliable, people-first content says age and low traffic alone are poor reasons to delete a page. Danny Sullivan, Google’s Search Liaison, and John Mueller, a long-time Google Search Advocate, have both said the same thing in public: refreshing or removing content purely to look fresh doesn’t move rankings the way people assume. The same caution applies, probably more once an AI model has already built part of its understanding of a brand on that page.
One more detail worth knowing, because it trips people up: noindex is a small instruction added to a page’s code that tells search and AI crawlers not to show it going forward. But it has real limits. It does nothing to erase what a model already learned from that page in the past. noindex manages what happens next. Not what already happened.
Don’t make this decision alone
Before retiring or rewriting anything at scale, check AI citation status first — not just traffic and age, the way most content audits default to running. That’s the single biggest lesson sitting inside the Sitecore example.
Past that, this isn’t a one-person call. Legal may have retention requirements. Web and development need to plan the redirects and migration. SEO has to weigh ranking impact alongside the AI-citation findings. Skip any of these, and a well-intentioned cleanup can turn into a bigger mess than the one it was trying to fix.
After the changes go live, repeat the same buyer-style prompts periodically and track how the AI tools’ answers shift — a separate signal from business outcomes like leads or traffic, since it tells you whether the correction landed with the AI model, not just whether the page looks fixed internally.
FAQs
Can deleting an old web page actually hurt how AI chatbots describe my brand?
Indirectly, yes. Deleting a page doesn’t erase what an AI model already learned from it. It only removes the chance to correct what that page said. Updating it in place is often safer than deleting it outright.
How do I find out if ChatGPT or Perplexity is already citing an outdated page on my site?
Ask the real questions a buyer would ask about your brand, across each tool, more than once, and save the answers along with any sources the tool names or links to. Then compare that list against your own pages — especially older ones about pricing, products, or positioning.
Does adding a “last updated” date fix an outdated page for AI search?
Not by itself. AI tools draw on what they learned during training as well as what they read live, so a date change alone doesn’t undo an outdated impression the model already formed. The content itself has to be accurate — not just the date stamp.
What does noindex actually do for a page AI tools have already learned from?
It tells systems that respect the instruction not to show or re-read that page going forward. It has no effect on what a model already learned from the page before noindex was added. It manages future access, not past training.
Who should be involved in deciding whether to retire an old page?
More than one team. SEO, legal, and web/development should all weigh in before pages are retired or redirected at scale — for ranking impact, retention requirements, and the technical migration itself.
However any one page shakes out, the bigger point holds: an AI model’s picture of a brand is only as current as the oldest page it still trusts. Checking that archive, honestly, is where real AI-visibility work has to start.