Why AI Recommends Your Competitor's Old Post, Not Yours

AI engines recommend a competitor's older content over your newer, better article because they choose sources by retrievability, how directly a page answers one specific claim, and whether other sources back it up. They do not compare two posts and reward the stronger one. If a rival's old page is easy to fetch, states the answer plainly and is echoed elsewhere on the web, it can keep winning the citation even when your post is more thorough.

This guide explains the mechanics using published research, gives you a diagnostic table to find which cause applies to your page, and walks through a seven-step fix. It replaces the common explanation that models simply "remember" older pages from training data, which is only a small part of the story for engines that search the live web.

Key takeaways

• Older does not mean favoured. An Ahrefs study of over 17 million citations found AI assistants cite content that is fresher on average than Google's top results, so age alone is not your competitor's advantage.
• Ranking and citation are different contests. Most links cited by ChatGPT, Gemini and Copilot are not in Google's top 10 for the same prompt, so beating a rival in classic search does not guarantee a citation.
• Diagnose per prompt and per engine. The cause on Perplexity (retrieval) is often different from the cause on ChatGPT without search (what the model learned).
• Fix three layers in order: access (can the engine fetch you), answer (does a passage state the claim cleanly), and corroboration (do other sources say the same about you).
• Retire or consolidate conflicting old pages on your own site. Yext argues stale and duplicate pages can count against a brand.
• Re-measure the same prompts on a fixed schedule so you can tell a real change from normal answer variation.

Why does an AI engine cite a competitor's older post instead of mine?

There are two separate mechanisms, and the old explanation blurred them. A model answering without live search relies on patterns learned during training, so a page that was widely quoted and linked before the training cutoff is more present in what the model absorbed. An engine that searches the web at answer time instead builds the answer from pages it retrieves, so what matters is whether your page is retrieved and whether a passage on it fits the claim being written.

The retrieval route is now the one most marketers see. OpenAI's publisher FAQ (checked November 2026) says any public website can appear in ChatGPT search, and that blocking OAI-SearchBot can keep a site out of summaries and snippets. In other words, a competitor can be cited simply because their page is reachable and yours is not.

Retrieval is also not a ranking contest in the way Google is. Answers are assembled from several sources, and Straight North (October 2026) describes an AI system breaking a prompt into smaller questions, gathering material from multiple sources and assembling a reply. That means your competitor's old page may win one sub-question (a definition, a number, a step) while your longer post wins none, because no single passage in it answers that sub-question cleanly.

Is the competitor's page really winning because it is older?

Probably not. The best public evidence points the other way: in the study reported by Search Engine News (August 2025), Ahrefs analysed over 17 million citations from leading AI tools and found they cite content 25.7% fresher on average than the pages ranked in Google's top results. Recency, if anything, helps a new page.

So when an old page keeps winning, age is a correlate of something else. Old pages have had years to collect links, mentions, quotes in newsletters and discussion threads, and to settle into a stable, unambiguous form. Those are the things to examine, and they are all things you can influence.

An older article is not preferred because it is old. It is preferred when it is reachable, states the answer in a quotable passage and is confirmed by other sources the engine also reads.

What do the published numbers say about ranking versus citation?

Three sourced findings frame the problem. Use them to set expectations before you spend on fixes.

A worked example using only those numbers

Take 100 links that ChatGPT, Gemini and Copilot cite across a set of prompts. On Ahrefs' average, 12 of them rank in Google's top 10 for the same prompt, which leaves 88 coming from pages that do not. If you only watch your Google rankings, you are looking at the source of roughly one in eight of those citations.

Now take Google's AI Overviews. In the July 2025 analysis, 76.10% of cited pages were in the top 10, 9.50% were between positions 11 and 100, and 14.40% appeared nowhere in the results. Applied to 1.9 million citations, 14.40% is about 273,600 citations from pages that did not rank at all. This is arithmetic on published figures, not a result for any one brand, but it shows why a rival's page can be cited without outranking you, and why the answer differs by engine.

How do the main AI engines choose between two similar pages?

Each engine mixes training knowledge, live retrieval and its own preferences differently. The table summarises what the sources above support, and where you should look first.

On Perplexity, Fokal (checked November 2026) describes sources as chosen for relevance to the specific claim rather than overall domain authority. That is a third-party analysis, not Perplexity documentation, so treat it as a working hypothesis and test it on your own prompts.

Which domains get cited also varies by engine and over time. Semrush (October 2025) analysed 230K prompts across ChatGPT search, Google AI Mode and Perplexity, and found Reddit and Wikipedia still among ChatGPT's two most-cited domains, with AI Mode showing a preference for Google's own domains. If your competitor's old page is the one that Reddit threads and encyclopedic pages point to, the engine is effectively reading corroboration, not just the page.

How do I find out why a competitor wins a specific prompt?

Use this decision table on each prompt where a rival is cited and you are not. Run the prompt on the engine in question, open the cited page, and match what you see to a row.

How do I fix it? Seven steps

Work through these in order. Each step is an action, the result you should see, and the reason it matters.

1. Build a fixed prompt set and record who is cited. Action: choose the questions your buyers ask, run each on each engine you care about, and log the cited URLs and brand names. Result: a baseline that shows which rival pages win which prompts. Why: without a baseline you cannot tell a real improvement from the normal variation in AI answers. See which prompts to track for AI search visibility and how many prompts you need.
2. Open the winning page and mark the passage being used. Action: read the competitor's cited page and highlight the sentence or table that matches the AI's wording. Result: a short list of the exact claims you must answer better. Why: engines cite at claim level, so you are competing for a passage, not a whole article.
3. Remove access problems. Action: confirm that OAI-SearchBot and other search crawlers can reach the page, and that the page is indexable and renders its text. Result: your page becomes eligible for retrieval. Why: OpenAI's publisher FAQ names blocking OAI-SearchBot as a reason a site may not appear in ChatGPT search. Use the AI crawler user agents list to decide what to allow.
4. Rewrite the opening of each section as a direct answer. Action: put a 40 to 60 word answer under every question-style heading, name the entity in full, and add the number or step the competitor's page offers plus something it lacks. Result: passages an engine can quote without surrounding context. Why: it removes the passage-level advantage the older page had. The sentence structure that gets cited guide shows the patterns.
5. Clear out conflicting pages on your own site. Action: list every page that states the same fact, merge duplicates, update old numbers, redirect or remove what is wrong. Result: one consistent version of each claim. Why: Yext (October 2026) argues AI checks a claim against many sources and that stale, duplicate or conflicting pages left live can count against a brand.
6. Add visible freshness and authorship signals. Action: show a real updated date, a named author and the sources you used, and change the date only when the content changes. Result: clearer provenance for both readers and engines. Why: freshness is associated with citation in the Ahrefs data, but a date change without edits is a trust risk. See how to add author and dateModified signals.
7. Earn corroboration outside your site. Action: publish original data others quote, get listed in the roundups and communities the engines already cite, and make your brand description match across profiles. Result: more independent sources saying the same thing about you. Why: the rival's older page has had years to collect these. Read how to publish original data to get cited by AI and Wikidata and sameAs for brands.

A checklist to run on the page you want cited

1. The page is public, indexable and not blocked for search crawlers.
2. The main answer is in text, not only in an image, tab or script-rendered widget.
3. Every H2 is phrased as a question a buyer would ask.
4. The first sentence under each H2 answers it directly and names the subject in full.
5. Each number has a named, linked source and a date.
6. One table compares options or steps, so a claim can be lifted cleanly.
7. Author, publication date and last meaningful update are visible.
8. No other page on your site contradicts this one.
9. Your brand description matches across your site and public profiles.
10. You have at least one independent mention of the page or brand in a source the engines cite.

What do brands get wrong when they chase an older competitor?

Waiting for the model to catch up

Waiting works only for answers that come from training data, and even then you cannot see when an update arrives. For engines with live search, the delay is in your own access, structure and mentions, which you can change today.

Rewriting the whole article instead of the passage

A longer, better article can still lose if no section answers the sub-question cleanly. Fix the specific passage first, then expand.

Changing the date without changing the content

Freshness is associated with citation, but a false update date harms trust. Update the substance, then the date.

Treating one prompt run as proof

AI answers vary between runs and between engines. Repeat each prompt several times on a schedule, as in a four-week ChatGPT test plan, and change one variable at a time as described in GEO experiment design.

Counting backlinks as the only corroboration

A mention in a thread, list or article that an engine already cites can matter more than another link from a page it never reads. Aim for the places the engines visibly cite, which you can see in the citations themselves.

Ignoring your own contradictory pages

Old product pages, retired pricing posts and duplicate guides all give engines conflicting claims. Audit them before publishing more.

How does Noma help you see which competitor pages win?

Noma by Nurdd is an AI search visibility dashboard that tracks whether a brand appears in answers from 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, polled daily. It records whether your brand is named and which page each citation comes from, shows citation frequency by category with a competitor comparison, and gives recommendations for improving AI visibility. For this problem, that means you can see which competitor page is cited on which engine for which prompt, instead of checking by hand.

Book a demo and we will walk through your own category: the prompts your buyers ask, which competitor pages the engines cite for them, and where your brand is missing. Noma also has a 3-day free trial at nomahq.app; a card is taken at signup and billing starts on day 3, so you are not charged if you cancel before then.

If you would rather have a team make the changes, Nurdd's generative engine optimization services and answer engine optimization agency work shapes the content and sources the engines read, with results measured in Noma. For background, read what generative engine optimization is, why your brand isn't showing up in ChatGPT answers and why your competitor is cited and you aren't. To benchmark rivals against you, see AI search share of voice.

What should you do this week?

Start with three prompts where a competitor is cited and you are not. Open each cited page, mark the passage that matches the AI's wording, then check access, answer structure and corroboration in that order. Make one change per page and re-run the same prompts on a schedule.

If your page is thorough but not quoted, the usual gap is a missing direct answer or a missing independent mention, not quality. Book a demo of Noma and see, for your own brand, which competitor pages the 13 engines cite for your category prompts and which of the seven steps to take first. You can also run the GEO readiness scorecard today, or read the guide to writing a brand description that engines can use.

How we researched this

We reviewed vendor research, platform help pages and industry publications on AI citation behaviour, content freshness and retrieval, and checked each linked source on the publication date of this post. We worked from the excerpts search returned for the top-ranking articles on this topic, and some full pages could not be retrieved, so we cite only figures that appeared in those excerpts. We did not run our own experiments, interviews or customer analysis, and the worked example is arithmetic on published figures, not a result for any brand. Third-party descriptions of how Perplexity ranks sources are labelled as such.

Frequently asked questions

Do AI engines prefer older content?

No evidence in the sources we reviewed says they do. Ahrefs' analysis of over 17 million citations, as reported by Search Engine News, found AI assistants cite content that is fresher on average than Google's top results. An old page tends to win because of links, mentions and a clear answer, not because of its age.

If I outrank a competitor on Google, will AI cite me first?

Not necessarily. Ahrefs found only 12% of links cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same prompt, while Google AI Overviews draw most citations from top 10 pages. Ranking helps most on Overviews and Perplexity and least elsewhere.

Does ChatGPT use training data or live search to answer?

Both, depending on the mode. With search, ChatGPT retrieves pages through OpenAI's search crawler, and OpenAI says blocking OAI-SearchBot can keep a site out of summaries and snippets. Without search, the answer reflects patterns from training.

How long until a new page gets cited?

We found no reliable published timeline, so we do not give one. Retrieval-based engines can in principle pick up a page once it is crawled, but whether it is cited depends on the passage and corroboration. Track the same prompts weekly to see when it changes.

Should I delete my old posts?

Only the ones that conflict with your current facts or duplicate a better page. Yext argues stale and conflicting pages can count against a brand, so update, merge or redirect them rather than deleting indiscriminately.

Do backlinks still matter for AI citations?

Links still help pages get found and ranked, which matters most for engines that lean on top results. For AI answers, independent mentions in places the engines already cite appear to matter alongside links, so aim for both.

Can I check which competitor pages AI engines cite for my category?

You can run prompts by hand and log the cited URLs, but it is slow and varies run to run. Noma by Nurdd automates this across 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, and records which page each citation comes from.

Is Noma free?

Noma has a 3-day free trial. A card is taken at signup and billing starts on day 3; if you cancel before then, you are not charged.