You do not outrank a competitor in AI answers, you get added to a citation set that already includes them, because there is no position one to take. Pew Research Center found 88 percent of AI summaries cite three or more sources and only 1 percent cite a single source, so the realistic goal in 2026 is joining the set rather than displacing anyone. The method: find the pages cited today, identify which source the model reads, publish something more specific on that question, then get referenced on the third-party domains it already trusts.
Why is displacement the wrong mental model?
Because the mechanics are different from blue-link search in three ways that change the whole approach.
There is no ranked list. An AI answer names two to six entities in prose and cites a handful of URLs. Being “second” is not a meaningful state. You are either named or you are not.
The answer is assembled per query, not retrieved from an index of positions. Ahrefs’ work on query fan-out describes how a single user question generates multiple hidden sub-queries, each pulling its own sources. Two people asking the same thing in slightly different words can get different citation sets.
And the source set is not only your competitor’s website. It is frequently a third-party page comparing you both: a directory, a trade publication, a review platform, a listicle. Ahrefs found that 67 percent of ChatGPT’s top 1,000 citations were on domains marketers cannot publish to. You often cannot win by improving your own page because your own page was never the source being read.
What is the method that actually works?
Five steps. The first two are diagnosis and most people skip them.
1. Capture the actual answer, repeatedly
Run the real query your buyer would type, in ChatGPT, Perplexity, Google AI Mode and AI Overviews, and record what comes back. Do it several times across several days, because outputs vary. Write down which competitors get named, which URLs get cited, and what specific claim each citation supports.
This is the baseline and it is the only thing that tells you what you are actually fighting. Firms that skip it end up optimizing a page nobody was reading.
2. Identify the source type, not just the source
Sort the cited URLs into categories. Competitor-owned pages. Third-party editorial. Directories and review platforms. Community content like forums and Q&A. Each category needs a different response.
If the citation is a competitor’s own comparison page, you can compete directly by publishing a better one. If it is a trade publication roundup, the move is earned media, not content. If it is a directory profile, the move is completing and updating your profile on that directory. The category determines the tactic, and this is where most effort gets misallocated.
3. Publish the more specific answer
Where the citation is a page you could have written, write the better version. Better here means concrete: named entities, current figures with sources stated in the text, a comparison table where the question is comparative, and a direct answer in the opening rather than three paragraphs of context.
The reason specificity wins is mechanical. A model assembling an answer needs extractable claims. A page saying “pricing varies by scope” offers nothing to lift. A page saying “retainers run $1,500 to $6,000 a month depending on whether earned media is included” offers a sentence that can be quoted.
Not sure which competitors are being named instead of you? Get a free visibility check and see the actual answers in writing.
4. Update relentlessly
Freshness is the most underrated lever in this whole category. Ahrefs’ 17 million citation analysis found AI-cited content averaged 25.7 percent fresher than organic results, that ChatGPT preferred URLs 393 days newer in in-text references and 458 days newer in citations, and that average time since last update for cited pages was 909 days against 1,047 for the organic results.
The counterintuitive finding in the same study: average age of cited pages was still 2.9 years, and Google’s AI Overviews actually cited content 16 days older than the organic listings on average. So freshness is a ChatGPT-weighted signal more than a universal one. Update your pages, but do not conclude that only new content gets cited.
There is a practical sequence inside this step. Update the pages that already get some retrieval before writing new ones, because a page the systems already read is a shorter path than a page they have never seen. Correct the modified date honestly when you make a substantive change, and leave it alone when you have only fixed a typo, since an inaccurate date on a page that did not change is the kind of signal that erodes trust in everything else on the domain.
5. Earn the third-party mentions
Where the citation is a domain you cannot publish on, the only path is being referenced there. Trade publications, industry roundups, directory profiles, review platforms. This is slow, it involves humans, and it is the part that cannot be automated, which is precisely why it is defensible once it works.
What does not move the needle?
Three things absorb budget and produce nothing measurable.
Structured data bought as a citation product is the biggest one. Ahrefs tracked 1,885 pages adding JSON-LD between August 2025 and March 2026 against 4,000 matched control pages: Google AI Mode came in at plus 2.4 percent, ChatGPT at plus 2.2 percent, both statistically indistinguishable from zero, and AI Overviews declined 4.6 percent relative to controls. A separate searchVIU experiment tested whether ChatGPT, Claude, Perplexity, Gemini and Google AI Mode used schema when fetching a page in real time and found none of them did, with every system extracting only visible HTML and ignoring JSON-LD, Microdata and RDFa. Schema is worth implementing for rich result eligibility, which Google documents. It is not the displacement mechanism.
Self-published “best of” lists placing yourself first are the second. Ahrefs studied 26,283 source URLs on this question specifically, and the pattern that emerges is that obviously self-serving lists carry less weight than independent ones. Writing a list where you win is transparent to readers and to the systems reading them.
Volume without specificity is the third. Publishing fifty pages on adjacent topics to “build authority” runs into Google’s scaled content abuse policy, which treats mass-produced content made primarily to manipulate rankings as spam subject to action. Ten pages that answer real questions with real numbers beat fifty that restate the category.
There is a fourth worth flagging because it wastes the most time: chasing the answer rather than the source. Teams watch an AI output name a competitor, then rewrite their homepage messaging to sound more like that competitor. The output is generated text, not a ranking you can reverse-engineer from wording. What produced it was a set of retrieved documents. Changing your prose without changing which documents exist and what they say leaves the retrieval layer untouched, which is why this feels productive and measures as nothing.
How do you measure whether you gained ground?
Four measurements, in descending order of reliability.
Direct answer capture is the most honest: the same queries, run on a schedule, recorded. Tedious and accurate. It tells you whether you are named, which no automated proxy fully replaces.
Citation tracking tools give you scale at the cost of precision. They sample queries rather than capturing everything.
Search Console’s generative AI features report gives impressions, and this needs a caveat stated plainly: it reports impressions, not citations, with no clicks column and no query breakdown. It is a directional proxy for visibility and treating it as a citation count will mislead you.
Referral traffic from AI platforms is real but small, and shrinking as a share of value. Pew found users who saw AI summaries rarely clicked the cited sources at all. Judging the channel by clicks measures the wrong thing, since the point is being named in an answer the user acts on without clicking.
Give any change two full quarters. Both content compounding and earned media operate on that timescale. For the adjacent question of appearing in Google’s summaries specifically, see how to rank in Google AI Overviews. If you want this run as an ongoing program, our services covers what that includes.
FAQ
Can you actually outrank a competitor in AI search results? Not in the way blue-link search works, because AI answers have no ranked positions. Pew found 88 percent of AI summaries cite three or more sources and only 1 percent cite a single source, so the realistic objective is joining the citation set alongside competitors rather than displacing them. You are either named in the answer or you are not.
How do you find out which competitors AI assistants recommend? Run the exact queries your buyers would type in ChatGPT, Perplexity, Google AI Mode and AI Overviews, several times across several days since outputs vary. Record which companies get named and which URLs get cited, then sort those URLs by type: competitor-owned pages, third-party editorial, directories, and community content. The source type determines what tactic can change it.
Does updating old content help with AI citations? Yes, particularly for ChatGPT. Ahrefs’ analysis of 17 million citations found AI-cited URLs averaged 25.7 percent fresher than organic results, with ChatGPT citing URLs about 458 days newer. The same study found average cited page age was still 2.9 years and Google’s AI Overviews cited content 16 days older than organic on average, so freshness is a real but platform-weighted signal rather than a universal requirement.
Will adding schema markup get me cited instead of a competitor? The controlled evidence says no. Ahrefs tracked 1,885 pages adding JSON-LD against 4,000 matched controls and found AI Mode at plus 2.4 percent and ChatGPT at plus 2.2 percent, both indistinguishable from zero, with AI Overviews down 4.6 percent. A searchVIU test found five major AI systems extracted only visible HTML during direct retrieval, ignoring JSON-LD entirely. Use schema for rich result eligibility instead.
Why do AI answers cite sites I cannot publish on? Because the models retrieve from whatever domains carry credible content on the question, and many of those are editorial, institutional or community sites. Ahrefs found 67 percent of ChatGPT’s top 1,000 citations sat on domains marketers cannot publish to. Where that is the case, the path is being referenced on those domains through earned media or a complete profile, not through publishing more on your own site.
How long before the work shows up in AI answers? Plan on two full quarters before judging. Content needs to be crawled, indexed and then retrieved, and earned media takes weeks to months to land and longer to be picked up. Answers also vary run to run, so a single check showing no change means very little. Measure on a schedule rather than reacting to one output.
The short version
Stop thinking about beating a competitor and start thinking about joining the set they are in. Capture the real answers, sort the citations by source type, publish the specific version of the page worth quoting, keep it current, and earn the mentions on domains you do not control. The tactics that feel like shortcuts, markup and volume and self-ranking lists, are the ones the data keeps failing to support. The slow ones keep working.
Want to see which answers name your competitors and not you? Start with a free audit and get the baseline on paper.
Tagged