What AI is actually reading before it recommends a vendor
The other 98% splits between two sources outside a marketing team's own output: independent review directories such as G2, Clutch, Gartner, and GoodFirms, at 47%, and community content led by Reddit, YouTube, and forum posts, at roughly 51%. Reddit alone shows up as a top-three citation source across nearly every industry, not only software.
Put simply, when someone asks an AI assistant to recommend a vendor in a given category, the model isn't summarizing that vendor's homepage. It's summarizing what other people said about it on a review site, or somewhere else on the internet the company doesn't control.
Why AI recommends vendors based on what customers say, not what brands publish
None of this is entirely new.
In a sense, this is just word of mouth wearing a new interface. Word of mouth was always the most trusted channel a brand had, and always the hardest one to manage directly, since it lived in individual conversations a marketing team could listen to but rarely shape. What's changed is that an AI model makes the conversation searchable and repeatable. It multiplies its reach, surfacing the same handful of reviews and threads to every buyer who asks a similar question. Which means a brand's standing on a small number of platforms now carries weight that used to be spread across thousands of scattered, one-off conversations.
Which is why review platforms, Reddit threads, and YouTube testing videos now shape the decision long before a vendor is ever contacted. Forrester's framing captures this well: it's an answer engine optimization question now, not simply a customer marketing nice-to-have.
What actually works over time
A handful of practices tend to show up wherever this is handled well, and they require consistency.
A steady stream of current reviews outperforms a single push. Companies that consistently earn reviews on the two or three platforms that matter most to their category get cited more often, and cited with more current information. A wave of reviews collected right before renewal season tends to read as exactly what it is: manufactured. It does produce a visible spike in volume, but both a buyer skimming the platform and, increasingly, the model summarizing it view them with distrust.
Responding or managing reviews? Both, neither alone is enough. Treating a negative review as something to manage quietly (maybe escalated internally, maybe eventually flagged for removal) simply leaves only one side of the story on the page. A reply gives the model something fresh and specific to reference.
But a reply without internal escalation doesn't fix the root cause: the same friction just reaches the next user. No one recommends a deficient solution. Responding and fixing are two halves of the same job, and they sit in different departments. Fix and communicate.
The work needs a clear owner. No single department can be responsible for what people say on Reddit or a review site. But someone needs to monitor and manage it constantly, on two fronts: the internal side (making sure issues are actually being handled, and that the fix can scale) and the external side (the public communication, and earning more reviews).
Presence on Reddit works only when it reads as a person, not a brand. Monitoring the subreddits relevant to a category and responding honestly, without turning it into a pitch, appears to be one of the more effective moves available right now. It also fails quickly the moment it reads as marketing in disguise.
Earned coverage travels further than announcements. Trade press built around original data or an honest comparison tends to get cited. A press release restating a company's own announcement usually doesn't.
The model is only repeating what customers already said
This is still a new discipline, but the pattern is already clear: it takes continuous care. The deeper mistake is a shorter time horizon than the problem calls for — treating visibility in AI answers as an isolated campaign instead of a continuous effort. A model summarizing hundreds of data points about a vendor doesn't need many outliers to tell recent, coordinated activity apart from a genuine pattern sustained over time.
There's no real shortcut underneath any of this. No content plan substitutes for a customer base genuinely willing to vouch for a company in public.
It's worth asking what an AI assistant would actually say if someone asked it to compare one vendor to another today. The answer has less to do with anything you published this year than with how the last few quarters of customer experience actually went.