When an AI Answer Cites a Page You Forgot to Update

Picture a prospect asking an AI assistant whether your business takes Saturday appointments. The answer says yes, from 9 to 1, and the footnote points to your website. You stopped opening on Saturdays last spring. The page the assistant read is an announcement post from three years ago that nobody took down. The assistant did what it was built to do: it found a source and cited it. The source was wrong because it was still up.

Most advice about getting cited by AI assumes the problem is absence. You aren't in the answer, so you publish more. A second problem gets less attention and is often cheaper to fix: you are in the answer, with a footnote, and the footnote points to something that stopped being true. An answer that names you with the wrong facts can do more harm than no answer at all, because the reader believes it and then finds out otherwise on the phone.

The free guide LLM SEO: The Complete Guide to AI Search Citation covers how language models decide what to cite and how to build the authority that earns citations. This post is about the step I would take before any of that: cleaning up the pages the assistants are already reading.

Why old pages keep getting quoted

Assistants that search the web while they answer draw on pages that are indexed and reachable, and age doesn't take a page out of that pool. An old promotion, a page for a service you discontinued, a staff page listing someone who left, a price list from years ago: if a search engine can still fetch it, an answering system can still read it. Files count too. Google lists PDF among the file types it can index (Google Search Central), and old brochures are some of the most stubborn sources on a small business site, because nobody thinks of a PDF as a page.

Old pages also tend to be specific. A post that says "open Saturdays 9 to 1" answers the question more directly than a current homepage that says "flexible scheduling". A system looking for a direct answer can prefer the stale, specific sentence to the vague, current one. That's uncomfortable, and it's also useful, because it tells you the fix has two halves: retire the old statement, and write the current fact as plainly as the old one was written.

Collect the footnotes first

Start by finding out which of your pages get cited, and for what. Write down the ten questions a buyer asks before contacting you: what it costs, when you're open, where you work, whether you handle a particular job, how long it takes. Ask each one in the assistants your buyers use, and open every source link that points to your domain or to a profile you control.

For each cited page, record four things: the question you asked, the URL, the claim the answer attributed to that page, and whether the claim is true today. The rows where the last column says no are your work list. The guide's 30-query tracking protocol is the fuller version of this exercise. For a first pass, ten questions are usually enough to surface the worst pages.

One run is a snapshot. Answers vary between sessions and between assistants, so ask the same questions again a week later before you decide a page is harmless.

Then search your own site for what changed

The footnote exercise finds the pages being quoted now. It misses the ones that could be quoted next month. For those, list every fact about your business that has changed in the past few years, such as prices, hours, locations, service names, staff, policies and phone numbers. Then look for the old version of each fact in four places:

• Search results. A site: search with the old wording in quotes, for example site:yourdomain.com "open Saturdays", shows indexed pages that still say it.
• Your CMS. Search the page and post archive for the old price or the old service name, including pages you thought were unpublished.
• Your media library. List every PDF and slide deck. A CMS search rarely looks inside them.
• Profiles you control. Directory listings and social bios often describe the business as it was on the day someone set them up.

Decide what each stale page becomes

Every stale page gets one of four treatments. Choose deliberately, because the choice decides what an answering system finds next time.

1. Update it. If the page still has a job, rewrite the changed fact and state the current one as plainly as the old one was stated. Put a visible last-updated date on it.
2. Redirect it. If a newer page does the same job, send the old address there. When a URL changes, Google recommends permanent server-side redirects such as 301 (Google Search Central).
3. Mark it as history. An announcement can stay up as a record, with a first line saying the offer or schedule has ended and a link to the current page.
4. Remove it. If nothing replaces it and it has no value as a record, take it down.

Then ask Google to recrawl what you changed. Its documentation points to the URL Inspection tool for individual pages and a sitemap for larger batches, and it says plainly that crawling can take from a few days to a few weeks and that a request guarantees nothing (Google Search Central). Other assistants refresh on their own schedules, which you can't control. Plan in weeks, and keep the footnote sheet so you can see when the answers catch up.

The sources you don't own

Some footnotes will point outside your site: an old directory listing, a partner's page, a review site, a local article from two years back. You can't edit those, but you can usually get them corrected. Claim the listings you haven't claimed. Send the partner the exact sentence that's wrong and the sentence that should replace it. For an article, ask the editor for a correction note on a factual error such as a closed location.

Keep expectations modest. A third-party page can take months to change, and some owners never reply. Your own pages are the ones you can fix this week, which is why I start there.

What this won't do

Cleaning stale pages doesn't make an assistant mention you more often. It makes the mentions you already get accurate. If your name is missing from the answers altogether, you need the other half of the work, which is becoming a recognizable entity and getting mentioned on sites the assistants trust. That work is slower, and it's most of what the guide covers.

I would still do the cleanup first. For a small site it takes an afternoon, and it stops an assistant from sending people to you with the wrong expectations. You also come out of it with a written list of the questions your buyers ask, which every later step needs.

Next step

Once the old pages are fixed, move on to earning citations on purpose. The free LLM SEO guide is a PDF download that walks through how models decide what to cite, the entity authority setup, the direct-answer content format and the 30-query tracking protocol for watching your citations over time. Start with the section on how models choose sources, since it explains why the stale page won in the first place.

← Back to all articles