Organic is down again and next quarter’s plan still funds links. Move that money onto lists. Ask ChatGPT, Google AI Mode, or Claude for the best software in a category and the engine builds its shortlist from “best X” roundups. So a place on a roundup it already cites does more to get you named than any backlink.
That holds for category and comparison questions in software, the “best,” “top,” and “alternatives to” questions where a shortlist gets made. It fails in a category too new for anyone to have written a roundup, where the job is writing the first one. And it says nothing about how-to or brand-name questions, which engines answer from other pages.
The engine turns the buyer’s question into several searches, and roundups are the pages titled for them. Each has already compared a dozen products, so the engine lifts names straight out of it and repeats the ones several lists agree on. A backlink only helps here by getting a page fetched, and on this question the page that gets fetched is the list.
.png)
Weigh that knowing who is saying it. Since 2020, Empact Partners has worked inside the marketing teams of B2B software companies, and getting partners onto these roundups is work we sell.
We call that work Existing Article Outreach, and inside our consultancy it sits in SEO, the workstream we call Search Everywhere Optimization because it covers every place a buyer searches, AI engines included.
Every Engine We Measure Builds Its Shortlist Out of Roundups
The Empact Panel is where we watch engines do this. It runs separately for each partner: every week it puts the same unbranded buyer questions to ChatGPT, Google AI Mode, and Claude, and stores every answer, every page cited, and every vendor named. In one B2B software category, the weekly runs on September 30 and October 7 produced 37,800 answers.
Even ChatGPT, the engine that leans on lists least, cites one in almost half its answers to these questions, and on Claude it is hard to find an answer without one. Each engine reaches the lists by its own route, and the routes explain why the format wins.
ChatGPT rewrites the question into searches lists are titled for
OpenAI’s own help page says ChatGPT search rewrites a question into one or more targeted queries, then sends narrower ones after reading the first results. Ask it which CRM suits a ten-person sales team, and the search that goes out can read “best CRM for small sales teams”: a roundup’s title, on a page built to rank for it.
ChatGPT also cites vendors’ own pages more than the other two engines do, and it reaches a list in fewer answers. When it does, the list does the naming. When ChatGPT cited a roundup the panel could open and read, 81% of the names it gave were on a list cited in that same answer.
In June, AIVO, a visibility vendor, split ChatGPT’s answers by what the person asking wanted and found the same reflex. ChatGPT cited a listicle in every one of its answers to people comparing options, in 13% of its answers to how-it-works questions, and in none of its answers to people looking for a particular site.
Google AI Mode fans one question out into many
Google says AI Mode breaks a question into subtopics and issues a multitude of queries at once. Each of those queries is another chance to land on a roundup. Three in four answers on AI Mode cited one, and lists made up 38% of everything it cited.
Lists do the least of the naming on AI Mode, even though it cites them more often than ChatGPT does. In answers citing a roundup the panel could read, lists were behind 65% of the vendors AI Mode named, and 28.5% had neither a cited list nor their own site behind them.
Google already knows a category’s established names, and Ahrefs’ study of 75,000 brands found AI Mode tracks brand signals more closely than ChatGPT does.
Claude reads the lists on almost every answer
Anthropic’s documentation says Claude can search again and again within a single request, and always cites what it used. On best-software questions, what it used was lists. 98% of Claude’s answers cited a roundup, and in answers citing a roundup the panel could read, 85% of the vendors Claude named were on a list it cited.
Claude also takes its lists from bigger publishers than the other two engines do. If a vendor’s own authority counted for anything, Claude is where it would show.
The same reasons hold on all three
Underneath the three routes, a roundup wins for the same reasons on every engine:
Every reason on that list belongs to the roundup. None of them improves because your site picked up another fifty referring domains.
Put Lists and Links Side by Side and the Lists Win on Every Engine
Qvery, our sister company, ran the first version of this comparison on ChatGPT in September. The comparison ranks the same brands two ways and asks how closely the two orders match, where 1 is the same order and 0 is no relation at all. Read it as a result from people who sell the measurement, and check the error ranges published with it.
The ranking closest to ChatGPT’s naming order, at 0.64, sorted brands by how often they sat on the roundups ChatGPT cited. Referring domains scored 0.38, a backlink-strength score 0.35, and organic traffic 0.27. On Qvery’s error ranges, lists beat the backlink score and traffic outright, and led referring domains by less than the margin.
We ran the same comparison on the panel, on all three engines, for the 50 vendors competing in that software category. For every vendor we counted the cited roundups that included it, engine by engine, looked up its domain rating, and set both against how often each engine named it.
The lists win on every engine, and on Claude, domain rating barely registers. The fair objection is that big brands have both: more links, and more lists that already include them. Lists and rating do rise together in this category, so we held one level and tested the other.
Hold domain rating level and the lists still line up with who gets named on every engine (0.60 on ChatGPT, 0.48 on AI Mode, 0.49 on Claude). Hold the lists level and domain rating lines up with nothing: −0.07 on ChatGPT, −0.05 on AI Mode, and a slight reverse of −0.26 on Claude. Rating bought naming only through lists.
Now split the 50 vendors twice, once at the median domain rating of 74 and once at the median number of cited lists. Most land high on both or low on both. The test is the 18 that land on opposite sides, and for the median vendor in each group, this is the share of the category’s answers that named it:
| Engine | Many cited lists, lower rating (10 vendors) | Higher rating, few cited lists (8 vendors) |
|---|---|---|
| ChatGPT | 16% | 5.8% |
| Google AI Mode | 23.5% | 10.7% |
| Claude | 39% | 21.8% |
The vendors with the lists and the lower rating were named nearly three times as often as the other group on ChatGPT, more than twice as often on AI Mode, and nearly twice as often on Claude. The groups are small, so read the table for direction, and the direction holds on every engine.
A vendor was named 55.8% of the time on ChatGPT when a cited list included it, against 11.7% when none did. Vendor by vendor, all 47 tested on ChatGPT were named more often where a list included them than where none did, as were 45 of the 46 on AI Mode and 35 of the 38 on Claude. A list raises the odds and decides nothing alone.
Google’s AI Overviews, the one engine our panel does not cover, point the same way. In Ahrefs’ study of 75,000 brands, a brand’s AI Overview mentions followed how often the rest of the web mentioned it far more closely than they followed its backlink count: 0.664 against 0.218, on the same scale.
Where a link still counts
Links matter on the publisher’s side. A roundup has to rank to be fetched, and its ranking runs on the publisher’s links. That is why a list that already ranks and is already cited is worth more than a new one.
We know what links do for Google because we built them for years in this workstream. Over three years with flair, one of our case studies, organic traffic rose 1600%, and the work included more than 500 links from sites rated DR 40 or above. The engines answering a best-software question read something else.
A roundup on a small site can still rank for the narrow searches an engine writes, and the citations show it. On ChatGPT, 43% of roundup citations came from sites rated below 30, the kind of site a link plan sorted by authority puts at the bottom:
Claude leans the other way, toward the big publishers, and the vendor’s own rating still says almost nothing about whether Claude names it.
.png)
Those numbers are the case for moving the budget. The case against it is serious, and it comes from people who watched the last generation of link tactics collapse.
Roundups Are Pay-to-Play, and the Engines Will Learn to Discount Them
Most of the objection is true:
Grant every point and a narrower case survives. The engines cite vendor-written lists today and name vendors out of them, as the panel records every week. An engine that discounts sponsored roundups still needs lists to answer a best-software question. The place on a list it keeps trusting is the one an editor gave because the product fits.
What a crackdown hits is volume. Thirty bought slots on thirty lists make a pattern anyone can trace, and we have watched a traced pattern get discounted before. In our link-building years it was guest posts, and the traffic fell off a cliff instead of sliding down a slope.
The cliff is why this work is slow and editorial by design. It aims at about ten of the hundreds of lists the engines cite in a category, and it earns each place. Buying a slot on every roundup in the category is the next link scheme, and we would expect it to end the way the last one did.
What We Do First, and What the Budget Buys Instead
Existing Article Outreach starts the way every partnership of ours does, with an audit and a roadmap, and here the audit is a census. The Empact Panel puts the partner’s buyer questions to all three engines every week and logs every cited page. The first deliverable is a list of the roundups the engines read, with the partner marked on or off each one.
We rank those roundups by how many answers cite them, and the ranking is steep at the top. In this category about 300 different lists were cited on each engine, and on ChatGPT the ten most-cited took a third of roundup citations. Those ten are the plan.
The places themselves come slowly. A senior consultant owns the partnership, and our team earns places on those pages one editor at a time, starting with lists that name competitors and leave the partner off. The panel shows each week whether naming has moved, and it takes quarters, because a list changes only when its editor refreshes it.
Count the Lists Before Anyone Signs Off Another Guest Post
Put five of your buyers’ questions to ChatGPT, Google AI Mode, and Claude, list every roundup their answers cite, and mark the ones that leave you off. That list is the off-page plan for the quarter, it comes before the next guest post, and the domain rating column can come off the sheet.
If organic is sliding and the engines keep naming competitors from lists you are not on, book a call with us, and we can find out together whether it is work we should do for you.
