Most of what an AI engine cites when a buyer asks which software to choose sits on someone else’s page: a ranked list, a Reddit thread, a review profile, an article under somebody’s name. In October, Google AI Mode put 72% of its citations in our own category on pages that none of the 164 companies we track there own.
Each of the seven ways onto those pages depends on a different person saying yes, from an editor to a subreddit to a platform you integrate with. So each entry says what the engines cite there, how a software company earns a place, and what it costs in time.
Empact Partners is a go-to-market consultancy for software companies, and we run this as Generative Engine Optimization from inside our partners’ marketing teams. We find the pages the engines read for a partner’s buyer questions, earn the product a place on them, and measure which of them start naming it.
| Way in | Who puts you there | Where it shows up most | The work, then the wait |
|---|---|---|---|
| Comparison lists | An editor or another vendor | Claude: 98% of answers | One ask per list, then the owner’s refresh |
| Review profiles | Your customers | ChatGPT and Google AI Mode: about 1 answer in 13 | A free claim, then reviews one customer at a time |
| Reddit threads | A subreddit | Google AI Mode: 35% of answers | An account with weeks of history, then days per thread |
| LinkedIn articles | Someone writing under their own name | Google AI Mode only: 28% of answers | One article, then the engine’s choice among a few |
| YouTube walkthroughs | A creator | Google AI Mode only: 7% of answers | A creator’s queue, then a short stay |
| Trade publications | A journalist or editor | About 1.5% of answers on every engine | Something worth reporting, then months |
| Integration directories | A platform you integrate with | ChatGPT: 5.3% of answers | A real integration, then the platform’s review |
Our Empact Panel supplied the engine numbers: 4,200 answers from ChatGPT, Google AI Mode and Claude to 500 questions about our own category, B2B marketing services, collected 4 to 23 October.
Two entries only show up in a software category, so their numbers come from Qvery, our sister company. Discount them knowing whose they are, and ask us for the queries behind them.
.png)
Get Into the Comparison Lists the Engines Already Read
Best for: categories where buyers ask for the best tool for a job, and every company whose buyers use Claude.
A comparison list is a page that ranks the products in a category: a “best payroll software for startups” roundup, a “vs” page, an “alternatives to” page. No other kind of outside page we sorted was cited more often, on any engine. A ranked or comparison list appeared in 98% of Claude’s answers, 84% of Google AI Mode’s and 63% of ChatGPT’s.
Over a third of the citations to lists, 36% on Google AI Mode and 40% on ChatGPT, went to a list published on a competitor’s own site. Those owners rarely add a rival, so the realistic targets are the independent lists and the ones written by companies that sell beside you rather than against you.
A list you are on can also be refreshed without you, and nobody tells you when it is.
Verdict: the first way in for almost every software company, and close to the only one that counts if your buyers ask Claude. It is also where our citation work for a partner starts, with the lists the engines read for the partner’s questions. Skip it only where nobody has published a list yet, and get an independent one written first.
Collect Reviews Where Software Buyers Compare
Best for: software whose buyers check what other users say before they shortlist anyone.
Review sites are G2, Capterra, TrustRadius and their neighbors: a profile per product, written by the people who use it. Across every brand Qvery tracked in August, most of them software companies, a review site was cited in 7.4% of ChatGPT’s answers and 8.0% of Google AI Mode’s, about one answer in thirteen.
In our own category the review sites barely register outside Claude, which cited them in 7.7% of its answers against under 2% for ChatGPT and Google AI Mode. Consultancies get reviewed in fewer places than software does. Either way, a profile with four reviews is a row in a grid with nothing in it worth quoting.
Verdict: worth doing for any software company whose customers like the product, because the claim is free and the work is asking. Skip paying for placement inside the site until the free profile has enough reviews to say something.
Answer the Reddit Threads That Are Still Open
Best for: buyers who use Google AI Mode or ChatGPT and want to hear from people who already bought.
Reddit is the outside page Google AI Mode cited most after the lists: in 34.8% of its answers, against 12.1% of ChatGPT’s and none of Claude’s.
The usual objection is that the cited threads are archived, so there is nothing left to join. On 28 October we checked the 448 cited threads Reddit would show us: 58% were more than six months old, and only 20% were archived or locked.
KKday shows the method at full stretch: three months, about twenty new threads a month, and a recommendation added only once a thread had earned attention. It came to 3,208,973 views and 60+ explicit KKday mentions, which are reach and mentions rather than citations.
Two things cut against it. Claude does not cite Reddit at all, and a new account’s first recommendation tends to be removed by the moderators.
Verdict: the strongest way in for Google AI Mode and a real one for ChatGPT, because 92% and 82% of their Reddit citations pointed at a thread you could still answer. Skip it if your buyers ask Claude, or if nobody on your side can answer a question without selling.
Get Named in the LinkedIn Articles People Write
Best for: B2B software whose buyers use Google AI Mode.
LinkedIn appeared in 31.2% of the answers from Google AI Mode and almost nowhere else: 0.1% of ChatGPT’s, none of Claude’s. The pages it cited were not posts. Of its 687 LinkedIn citations, 602 went to articles, the long pieces people publish under their own name, and 38 went to posts.
Qvery’s August data splits the same way, with articles in 5.9% of Google AI Mode answers and posts in 2.0%. Google AI Mode also settled on a handful of articles and cited them again and again, so a new one competes with an incumbent the engine already uses.
That concentration is the first catch. The second is trust: a list on an employee’s profile that ranks the employer first is an advertisement with a byline, and a reader can tell.
Verdict: worth it for B2B categories where buyers use Google AI Mode, and close to worthless for ChatGPT and Claude. Skip company-page posts for this: Google AI Mode cited company pages 41 times in 2,000 answers.
Get Walked Through on YouTube by Someone Else
Best for: products that are easier to show than to describe, sold to buyers who search on Google.
Google AI Mode cited YouTube in 7.1% of its answers, and ChatGPT and Claude never did. About a quarter of the answers that cited it pointed at a video promising a how-to, a demo or a review in its title.
That turnover is what a video costs you: the lists and review pages the engines cited held much steadier between collections, so a video buys a shorter stay than a list does. Google’s other AI surface leans on video harder. In Otterly’s study of 100 German queries, YouTube took 6.33% of AI Overviews citations against 2.02% in AI Mode.
Verdict: a second-wave move for Google-heavy categories with a product that demos well. Not for categories whose buyers ask ChatGPT or Claude, where we did not see a video cited once.
Give Trade Publications Something to Report
Best for: companies with news or original data a journalist would report.
Trade publications and the business press feel like the obvious place to be seen, and the engines barely cite them on category questions: in 1.7% of ChatGPT’s answers, 1.5% of Google AI Mode’s and 1.5% of Claude’s. What did appear was mostly market news, such as falling clicks or a regulator’s ruling.
The knock is the ratio: months of work for a page that category answers cite about as rarely as any other way in. It earns its place downstream. Wikipedia writes about companies only once independent press has covered them, and Claude cited Wikipedia in 14% of its answers.
Verdict: fund it for market questions and for the independent coverage Wikipedia requires, never as the route to a category answer. Leave it alone as a citation tactic if all you have to announce is a feature.
List Your Integrations Where Buyers Look Them Up
Best for: software that connects to tools its buyers already run.
Integration directories are the app pages on Zapier, HubSpot’s marketplace, Salesforce AppExchange, Shopify’s app store and Slack’s directory. Our panel cannot see them, because a consultancy integrates with nothing. Qvery’s August data can: one was cited in 5.3% of ChatGPT’s answers and 3.8% of Google AI Mode’s.
Zapier also fixes the first line of every listing: the app’s name, then “is a”, then what it does, which is a sentence an answer can lift whole. No outreach version exists, though. Only a real integration qualifies, and the listing describes you on the platform’s terms.
Verdict: the cheapest citation an integrated product leaves unclaimed, and more of a ChatGPT surface than a Google one. Wait if the integration does not exist yet: build it because customers need it, and the listing follows.
Which Way In to Fund First
Start from the engine, because each one has already chosen the pages it reads.
What our numbers cannot tell you is how long a placement takes to be cited. The panel shows which pages each engine reads, not how fast a new one gets in, and a timeline quoted for your category without that measurement is a guess.
When a partner brings us this problem, we start with an audit: their buyers’ questions put to all three engines every week and every cited page logged, so the roadmap names the sites worth a mention page by page. A named senior consultant owns it, our specialists run the outreach and the Reddit work, and the momentum takes months.
If your competitors keep turning up in AI answers on pages you have never been near, book a call with me, and let’s see whether we could work together.
