Before You Delete Old Posts, Check If AI Still Cites Them
You have 80 old posts, your Google traffic dropped, and someone told you to delete everything that pulls fewer than a handful of visits a month. Before you do, check whether ChatGPT or Perplexity is using those pages. A post that looks dead in Search Console can be the one an AI assistant reads every time a customer asks about your service. The audit guides ranking at the top of Google for this question score each page on Google traffic, backlinks and link equity (the ranking power a page passes on through its links). Then they hand you a delete list, and that default now throws away pages that are still working. This article gives you a page-by-page method for scoring both surfaces, then sorting every URL into keep, fix, or kill.
Your traffic drop is not proof the page stopped working
Search did not shrink while your clicks did. Comscore's Q2 2026 data was reported by Danny Goodwin at Search Engine Land on September 22, 2026. It puts US desktop search at 77 billion queries for the quarter, up 8% from 71 billion in Q2 2024. Americans are running more desktop searches than they were two years ago. If your posts lost traffic, "search is dying" does not explain it, and you should not plan around it.
The answers moved. The same Comscore report found AI Overviews, the AI-written answers Google places above its results, on 39.4% of US desktop searches in June 2026, up from 25.8% in July 2025.
Two in five US desktop searches now show an AI Overview. An informational post that used to earn the click can keep ranking in the same spot, stay just as accurate, and still lose visitors. The page itself did not change. The screen around it did.
That makes your Search Console chart an incomplete witness. It measures one surface, Google's blue links. It cannot see the second surface, the assistants like ChatGPT and Perplexity that read your pages, pull the substance out, and hand it to your customer. Sometimes your name comes attached and sometimes it does not. Delete a page based on the first surface alone and you are making a money decision with half the evidence.
I run this two-column audit on my own site. The content command center I built for my business has an archive audit that flags weak and orphaned posts (posts no other page on the site links to). It also has an AI visibility tracker that logs when assistants mention or cite my pages. The same command center's link audit took orphan posts from 14 to 1 across a 47-post archive. The site now has 100 published posts, and no page there gets cut on one column of evidence.
The top-ranking audit guides leave out the second column: a per-page check of whether AI engines cite the page. The rest of this piece shows you how to fill that column in yourself, without a paid tool, before a single URL gets deleted.
A one-screen audit deletes pages that AI is still reading
The standard audit method has not changed since 2019. Export every URL, sort by sessions, draw a line, and everything below the line gets redirected or deleted. The logic held up when Google was the only place your content got read. It does not hold up in 2026, and Jon Nastor of Digital Commerce Partners spells out why in Search Engine Land's content strategy guide.
Read the last sentence again with your own delete list in mind. A page with zero backlinks and falling Google clicks scores badly on every metric a traditional audit tracks. An assistant does not care about the backlinks. It cares whether the page answers the question with depth and credibility. The two surfaces can give the same page opposite grades, and the traditional audit only reads one of them.
Picture a physio clinic in Manchester with 80 blog posts written between 2019 and 2023. One of them explains how long sciatica usually takes to settle and when to stop waiting and book an assessment. Google sends it a trickle of clicks now, because an AI Overview answers the question first. On the agency's spreadsheet it is near the bottom, flagged for removal. Meanwhile a patient in Salford types almost that exact question into ChatGPT, and the answer she gets is built partly from that post. Delete it and the clinic loses the one page doing its explaining in a place its patients actually ask. The spreadsheet shows no loss, because the spreadsheet never measured the gain.
The opposite mistake costs money too. Search Engine Land reported on September 24, 2026 that Google's September 2026 spam update is rolling out worldwide over up to about two weeks, aimed at scaled and thin content. Pages nobody reads on either surface are the likeliest kind of thin content, and thin content is what that update targets. The "keep everything, just in case" plan fails as well. You need a way to tell a useful low-traffic page from a dead one, and traffic alone cannot do that.
AI engines use content more often than they credit it
The hardest part of scoring the AI surface is that the evidence hides. Comscore measured this in the lodging category from December 2025 to May 2026. It compared how often a brand's content was used as a source for an AI response with how often that brand appeared as a visible citation.
| Brand | Used as a source in AI responses | Shown as a visible citation |
|---|---|---|
| Tripadvisor | 61% | 21% |
| Hotels.com | 50% | 18% |
| Booking.com | 33% | 9% |
Tripadvisor fed 61% of the AI answers in that category and got a visible citation in 21%, a 40-point gap between being used and being credited. Booking.com's rates are smaller and the gap proportionally wider: used in a third of answers, visibly credited in fewer than one in ten.
For your audit, a check that only looks for a clickable link will undercount how much work your pages do. If you ask ChatGPT a question and your URL does not appear, the page is not automatically off the hook for deletion. Look for your brand name, your specific advice, your phrasing, your prices. Those count as use even when they come without a link. Comscore measured big travel brands, not clinics or contractors. The exact ratios will not carry over to your site. The pattern does: use outruns credit.
The second hidden cost shows up on the far side of a citation. Nastor gives it a name: Ghost Rankings.
"It's what happens when ChatGPT or Perplexity cite your content and recommend your brand, but then suggest a competitor for the actual purchase or hire."
Jon Nastor, Digital Commerce Partners, "Content strategy in 2026," Search Engine Land guide.
A page stuck in Ghost Rankings is not a keeper just because it gets cited. It is doing your competitor's education for free. That page belongs in the fix pile, with a sharper answer, your own proof, and a reason to hire you instead of the name the assistant suggests next.
Nastor's guide also gives the sorting logic, in three lines: "Performing on both surfaces → maintain and protect / Google down, LLM visibility up → enhance / Underperforming on both → cut or consolidate." LLM visibility just means whether the AI assistants use you. His three lines do not cover the page Google still loves that AI ignores. The third row of the matrix below is my own call. And his guide, like the top-ranking audit guides, gives you no per-page method for filling in the AI column in the first place. That method is the next section, and it is the piece your current audit is missing.
| AI assistants | Verdict | What you do with the page | |
|---|---|---|---|
| Still sends visitors | Uses or cites it | Keep and protect | Leave the URL and structure alone. Update facts and dates. Do not rewrite what already works. |
| Falling | Uses or cites it | Enhance | Sharpen the answer, add your own proof, and give the reader a reason to hire you, not the competitor the assistant suggests. |
| Still sends visitors | Ignores it | Keep, and make it quotable | Put a plain two-sentence answer near the top and add a short FAQ so that an assistant has something clean to lift. |
| No visitors | Ignores it | Cut or consolidate | Move any useful paragraphs into a stronger page and redirect the old URL there. |
Score every URL on Google and on AI before you touch it
You can run the whole audit from a spreadsheet, Google Search Console, and the same chat assistants your customers already use. On an 80-post site it takes a few focused sessions, and the order matters more than the tooling.
Step three is where most audits skip ahead, and it is the step that decides whether you delete a working page. Search Engine Land documented a manual version on September 24, 2026, in its piece on nine checks to audit AI visibility without a paid tool. It says to ask the assistants your customers use the questions they would ask, check whether your pages get cited, and check for brand mentions. My version per page looks like this.
If you want the fuller version of that exercise for your business as a whole, not page by page, I walk through it in how to check whether AI recommends your business.
Do it yourself: one person, one spreadsheet, and the pages sorted by Google clicks so you check the weakest ones first, since those are the ones in danger. Delegate it: hand a junior team member the checklist above and ask for the sheet, not a summary, because the verdicts are yours to make. Or, if an agency is running your audit, ask them one question before you approve anything: "Which column in your sheet shows whether ChatGPT or Perplexity uses this page?" If they have no answer, the delete list is not ready, however polished the deck looks.
Once the fix pile is sorted, you still have to decide which of those pages deserve your limited writing hours before any new content does. That effort question has its own answer, covered in when to refresh old posts versus publish new ones. This audit only tells you which pages are alive. That piece tells you where to spend the time.
Judge the audit by what the kept pages earn, not by how many you cut
An audit that ends with "we removed 40 pages" feels like progress. It is a count of deletions, and it says little about whether the business is better off. The measure that matters comes from the pages you kept and the pages you enhanced.
Nastor's line cuts both ways. Yes, you have more pages than you need. But authority is the scarce thing, and a page an assistant already trusts enough to use is authority you already have. Throwing it away to make the archive tidier trades the scarce asset for the cheap one.
Track these instead, on this schedule. Re-run the AI check on your enhanced pages one quarter after you change them, using the same questions you logged the first time. Note whether your brand now appears in the recommendation and not only in the explanation. That is the Ghost Rankings test: cited and chosen, or cited and passed over. Check Google clicks on your kept pages over the same window, comparing them with the same months a year earlier so seasonality does not fool you. Watch inquiries and bookings that mention the topic of a page you enhanced, because that is where a page earns its place.
Traffic was never the right final grade for a page anyway. If you want to judge your content by the revenue it drives, and not by visits, that is a separate method, laid out in measuring content value in revenue instead of Google traffic. The two-surface audit decides what stays. The revenue view decides what each page is worth.
Seventeen years in search have taken me across more than 300 businesses, from clinics to ecommerce to fintech. They have made me warier of the page deleted on one report than of the page kept a quarter too long. A kept page can be cut next quarter with no loss. A deleted page stops teaching the assistants the moment it goes, and no chart shows you what that cost. In 2026 there are two reports on every page, and most audits still read only one.
Frequently Asked Questions
Should I delete old blog posts that get no Google traffic?
Not until you have checked whether AI assistants use them. A post that gets almost no clicks from Google can still be the page ChatGPT or Perplexity reads when a customer asks about your service. Ask the question that post answers in the assistants your customers use, and note whether your page or your brand shows up. If it gets no Google traffic and no AI mentions, merge the useful parts into a stronger page and redirect the old URL. If AI is using it, improve it instead of deleting it.
How do I check if ChatGPT is citing my website?
Type the questions your customers would ask, in their words, into ChatGPT, Perplexity and Google, and read what comes back. Record whether your page appears as a linked source, whether your brand or your specific advice is mentioned without a link, and which business the answer recommends for the actual purchase. Run each question on more than one day, because answers change between sessions. You do not need a paid tool for this; Search Engine Land published a manual nine-check method on September 24, 2026.
Will Google's September 2026 spam update hurt my site if I keep old thin posts?
Search Engine Land reported on September 24, 2026 that the update is rolling out worldwide over up to about two weeks and targets scaled and thin content. The target is thin content, not old content. An older page that answers a real question well is not the problem. The pages most exposed are the ones dead on Google and never used by AI, because those are the likeliest to be thin. Merge or cut those first, and redirect each URL to the closest page you keep.
Your delete list was built from one screen, and that screen cannot see the second place your customers now ask their questions. A page with no clicks is not the same as a page with no readers. Before you sign off on the cut, fill in the AI column for every row, even if it takes an extra week. If you would like a second pair of eyes on your keep, fix and kill sheet before anything comes down, book an AI visibility call and we will go through your borderline pages together.
