Robby Stein at Google announced on August 26 that AI Mode link carousels are rolling out for developing topics, following the same feature landing in AI Overviews earlier this year. Coverage framed it as Google linking out more, which is true and also the least interesting thing about it. The more useful read is that the citation slot most GEO programs have spent two years optimizing for is now one of five, and the other four qualify on criteria that program never touches.
What AI Mode link carousels actually are
When a query concerns a developing topic, a trending story, an unfolding event, something the index is actively churning on, AI Mode now surfaces a prominent carousel of clickable article links alongside the generated answer. It gives preference to a user's Preferred Sources where they have set any. Stein described the intent plainly: the goal with AI search is to make it easy to connect with original coverage and a range of perspectives.
The mechanic that matters is the trigger. This is not a general-purpose link block appended to every answer. It fires on developing topics, which means the qualification is partly about recency and partly about whether Google classifies the query as actively developing at that moment. A piece published four days after a story breaks does not compete for this slot no matter how good it is, because by then the query has usually stopped being classified as developing.
That makes it the first AI citation surface where publishing speed is a genuine ranking input rather than a nice-to-have. It also makes it the surface most B2B content operations are structurally worst at, because a two-week editorial cycle with three review stages is the opposite of what qualifies.
Set expectations honestly about what it returns. Search Engine Land's coverage of the rollout makes the sober point that more link surfaces should mean more clicks from AI Mode, and will not restore pre-AI traffic levels. Nobody at Google has suggested otherwise. The realistic framing is that these slots decide how a shrinking pool of clicks gets distributed, not that the pool is refilling. Winning slot share in a smaller market is still worth doing, and it is a different objective from recovering 2023 sessions.
Citation stopped being one thing
Put the surfaces side by side and the fragmentation is obvious. Four of these five were announced or expanded within the last four months, and they qualify on genuinely different signals.
| SLOT | WHERE IT APPEARS | WHAT QUALIFIES YOU | ANNOUNCED |
|---|---|---|---|
| Inline source link | Inside the generated answer | Extractable structure, demonstrated authority, machine access | Ongoing |
| Link carousel for developing topics | AI Overviews, and AI Mode as of August 26 | Timely coverage while Google classifies the query as developing | August 26, 2026 |
| Preferred Sources | AI Mode and AI Overviews responses | A user has manually selected your publication | May 27, 2026 |
| Perspectives carousel | AI Mode and AI Overviews, developing topics | Timely articles plus forum and social discussion | May 27, 2026 |
| Highly cited label | Web article links across Search, not only AI surfaces | Being the primary source other published articles reference | May 27, 2026 expansion |
Duncan Osborn, a product manager on Google Search, announced the May 27 group. Google reported more than 345,000 unique sources selected globally as Preferred Sources, and said users are twice as likely to click through to one. That 2x figure is the most commercially interesting number Google has published about any of these surfaces, and it is attached to the slot that requires the least algorithmic work and the most audience work.
The highly cited label is the odd one out in that table and the most instructive. It is not confined to AI surfaces at all: Google applies it to web article links across Search, flagging pieces that other published articles frequently reference so readers can find the primary reporting behind a story. It is a machine-readable statement about who originated a claim, which is a signal search engines have always used implicitly and are now willing to show to users directly. For a company that publishes original research, that is the closest thing on this list to a durable moat, because the qualification cannot be reverse engineered by anyone who did not do the underlying work.
“Four of these five slots reward something other than page quality. That is the part a page-level GEO checklist cannot fix.”
Why most programs compete for only one slot
The inline source link is the slot everyone builds for, and there is a good reason. It is the one that responds to work a content and technical team can do alone: structure the page, mark it up, make the claim extractable, keep the crawler unblocked. That is the whole substance of how an AI citation gets earned, and it is real work with a real return.
The problem is that it is also the only slot where page-level optimization is the binding constraint. The other four are gated on things that are not properties of a page at all.
There is a budgeting consequence buried in that list. A GEO program funded entirely out of the content and technical line can only compete for one and a half of these five slots. The other three and a half need editorial operations time, communications time, and named subject experts with permission to speak publicly, none of which typically sits in an SEO budget or reports to whoever owns the visibility number. When a quarterly review asks why AI visibility has plateaued despite everything on the technical checklist being green, this mismatch is usually the answer, and it is an organizational finding rather than a tactical one.
None of that means the page work stops mattering. It means the returns on more page work flatten once a page is genuinely well structured, and the next increment of visibility comes from a different budget line. Teams that keep spending the whole GEO budget on page optimization are often optimizing a slot they already win.
The forum slot nobody staffs
The perspectives carousel deserves its own paragraph because it is the one enterprise teams reliably ignore. It pulls in forum and social discussion alongside published articles, which means a well-argued thread in the community where your category actually talks can occupy a slot your own site cannot reach on that query.
That is consistent with what we have seen across engines for two years now. Community and first-person expertise content keeps clearing bars that polished vendor content does not, which is why community sources kept turning up in our work on the link types AI engines trust. The uncomfortable version for a brand is that you cannot buy this slot and you cannot ship it from your CMS. Somebody who works at your company has to be a credible participant in a public conversation, under their own name, before the question gets asked.
For most organizations that is a staffing and permissions problem rather than a marketing one, which is exactly why it stays unsolved. It is also why a deliberate presence in the threads engines read tends to outperform its cost by a wide margin in categories where the discussion is concentrated in a handful of communities.
Where AI Mode link carousels fit in a quarterly plan
Do not restructure a program around a feature announced yesterday. AI Mode is still a small surface by search volume, and Google has not said which query categories will trigger carousels most often or how the classification of a developing topic is drawn. Treat this as a signal about direction, not as a mandate, and size the response accordingly.
| SLOT | EFFORT TO COMPETE | WHO OWNS IT | WORTH FUNDING NOW? |
|---|---|---|---|
| Inline source link | Moderate, mostly done once | SEO and content | Yes, and most teams already do |
| Highly cited label | High, requires original data or reporting | Content and research | Yes, it compounds across every surface |
| Perspectives carousel | Moderate, requires named people posting | Comms and subject experts | Yes in community-heavy categories |
| Preferred Sources | High, audience work with a long payback | Brand and lifecycle | Only if you already have an audience motion |
| Developing-topic carousels | Low to moderate, an editorial speed change | Editorial operations | Test with one fast lane, do not rebuild |
The row worth arguing about is the highly cited label. It is the most expensive to compete for and the only one whose qualification, being the source other people reference, also improves your position in all four other slots and in classic organic at the same time. Original data is the asset with the widest blast radius here, and it is the reason we keep pushing clients toward earned media over lighter-touch source plays when the budget is genuinely constrained.
What we would change this week
Two things, both small. First, take your existing AI visibility tracking and stop recording appearances as a single binary. Record which slot the appearance came from. Right now most tracking answers did we get cited, when the useful question is which of five slots did we get, because the answers point at completely different teams. That takes an afternoon of taxonomy work and it changes what the next quarterly review argues about.
Second, identify whether any topic in your category ever gets classified as developing. For a lot of enterprise categories the honest answer is rarely, in which case carousel slots are not your fight and you should say so out loud rather than quietly adding a fast-publish workflow nobody will use. For categories tied to regulation, security incidents, funding cycles or platform changes, the answer is often weekly, and a single approved fast lane for one editor is a cheap experiment with a clear read on it.
The broader pattern is worth holding onto beyond this particular feature. Google spent the first phase of AI search compressing many links into one answer, and it is spending this phase adding structured places for links to come back, each with its own gate. Visibility is becoming a portfolio of qualifications rather than a single ranking, which is closer to how earned media has always worked than how search used to. Programs built as link building and digital PR with a technical layer are adapting to that faster than programs built as pure page optimization, and in crowded categories like B2B SaaS the gap between those two shapes is already visible in who gets named.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.