Google AI Overviews cite pages that are already indexed and already eligible to appear with a snippet, and among those they favour pages answering the exact sub-question the summary is assembling. The reason some brands appear over and over is usually not that their own website is better. It is that other people write about them.
That second sentence is worth sitting with, because it points the work somewhere most content budgets never go.
Start with what Google says, because it is unusually plain
There is no separate algorithm to court and no markup that buys entry. Google states that to appear as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown with a snippet, and that there are no additional technical requirements. It also says there is no special schema.org structured data you need to add.
Take that literally. If you are not indexed, nothing else matters. If you are indexed and never cited, the cause is not a missing file, and anybody selling you one is selling you a story.
The filter to apply first. Before asking why an Overview ignores you, check whether the page ranks at all for the underlying query. Overviews draw on the ordinary index. A page nobody can find in normal search is not being snubbed by AI, it is absent from the pool being drawn from.
The finding that reframes the question
A 2025 study by Mahe Chen, Xiaoxuan Wang, Kaiwen Chen and Nick Koudas ran controlled experiments across ChatGPT, Perplexity and Gemini and reported that AI search shows a systematic and overwhelming bias towards earned media, meaning third-party authoritative sources, over brand-owned and social content.
Read that as an instruction rather than an observation. Your own pages compete in the category these systems trust least. An independent comparison or a well-regarded industry write-up carries more weight on the same claim than your homepage does, and no amount of rewriting the homepage changes which category it belongs to.
It also explains why the effect compounds. A brand that gets written about gets cited, being cited makes it more likely to be written about again, and the distance between the named brands and everybody else widens without either side changing anything.
| What you control | What it can achieve | What it cannot |
|---|---|---|
| Your own pages | Being used as the source for a specific factual claim | Reliably being named as the recommended option |
| Your structured data | Making clear which company the name refers to | Adding weight to a claim about how good you are |
| Independent coverage | Being named as a choice, repeatedly | Anything quickly |
| Reviews and comparisons | Shifting which set you get grouped with | Working if the underlying product view is poor |
Worth separating before you spend
Being absent from Overviews and being absent from the index underneath them are different problems with very different price tags, and telling them apart takes about twenty minutes. Our free AI visibility check reports both, including the cases where the answer is that ordinary SEO is your constraint.
The citation is worth less than it looks, and still worth having
Appearing inside an Overview does not send much traffic. Research tracking nearly 69,000 Google searches found that when an AI summary appeared, users clicked a source cited inside it on around 1 percent of visits.
So the value is not the click. It is that your name sat in front of somebody assembling a shortlist, which is a different kind of win from a visit and has to be measured differently.
What separates the brands that keep appearing
Four properties show up repeatedly in the pages that get pulled into summaries, and none of them are exotic.
- The page answers the sub-question rather than the topic. An Overview assembles its answer from several narrow claims. A page covering everything about a subject often answers none of those claims cleanly enough to be lifted out.
- The claim is checkable. A number, a threshold, a named condition. An assertion nobody can verify is cheap to leave out and risky to include.
- The company resolves to a distinct thing. Where a name is ambiguous, naming it carries risk, and hedging costs the system nothing at all.
- Somebody else says it too. A claim corroborated away from your own domain is a claim worth repeating.
The first two are editing work. The third is entity work and is mostly a week of unglamorous fixes. The fourth is slow, and it is the one that decides whether you get named rather than merely quoted.
What we would actually do
Check indexation first, because the rest of the question is moot otherwise. Then take the ten questions your buyers ask before they know you exist, and look at whether any page of yours answers one of them in a single liftable paragraph. Usually one or two do, by accident, and those are worth studying rather than replacing.
After that the work leaves your website, which is an uncomfortable conclusion for a content budget to reach. Getting into the comparisons and independent write-ups that already get cited in your category is slower than publishing, and it is the part that moves recommendations.
If a proposal you are reading describes nothing happening off your own domain, it is aimed at citations rather than at being chosen. Worth knowing which one you are buying.
Which third-party sources actually carry weight
Earned media is a broad term and it is worth narrowing, because the categories do not carry equal weight and some of them are far cheaper to reach than others.
| Source type | Why it carries weight | Effort to reach | Honest caveat |
|---|---|---|---|
| Review platforms | Structured, comparative, and already heavily cited in software categories | Low, and mostly process | Volume without recency reads as stale |
| Independent comparisons | Written to be neutral, so the claim survives scrutiny | Medium, and not fully controllable | You may not come out ahead |
| Industry publications | Editorial standards make the claim safe to repeat | High, and slow | Rarely responds to outreach alone |
| Community discussion | Unprompted, which is precisely why it counts | Cannot be manufactured honestly | Attempting to manufacture it usually backfires |
| Integration and partner docs | Third-party domain, factual, often overlooked | Low, if the partnerships exist | Only covers what you integrate with |
That last row deserves attention. If your product integrates with anything, the other party frequently documents the integration on their own domain, and that page is third-party, factual and permanently online. Most companies never check whether those pages exist, whether they are accurate, or whether they describe the product as it is now.
Topic pages lose to sub-question pages
The mechanical reason a sprawling guide underperforms is worth seeing rather than describing. An Overview answering a buying question assembles several narrow claims. Consider what it needs for a question about contract software for small legal teams.
It needs a definition of the category, a threshold where manual work stops scaling, a note on pricing structure, and a sense of which tools suit which team size. Four claims, potentially four different sources. A single long guide covering all four in flowing prose supplies none of them cleanly, because each claim is entangled with the sentences around it.
A short page answering one of those four in its opening paragraph supplies exactly one, cleanly, and gets used. This is why thin pages sometimes beat thorough ones here, and it is not a reward for thin content. It is a reward for a self-contained answer.
The trap that wastes the most money
There is a recurring pattern of buying tactics aimed specifically at Overviews. Special markup, AI-oriented files, question-and-answer rewrites of entire sites. Google has stated plainly that no special structured data and no new machine-readable files are required, so these are being sold against a documented position rather than an uncertainty.
The reason the pitch works is that it offers something inside your control at a moment when the honest answer involves other people. Rewriting your own pages feels like progress and can be scheduled. Earning independent coverage cannot be scheduled and often fails.
That does not make the owned work pointless. It makes it the smaller half, and worth sizing accordingly rather than treating it as the whole programme.
How to tell whether the earned work is doing anything
The difficulty with off-domain work is that it produces no dashboard. There is no equivalent of a rankings report, and the lag is long enough that people abandon it before it shows.
Three things are worth tracking, and all three are free. Count how often your name appears when you ask category questions across several engines, recorded as a fraction with the denominator written down. Watch branded search volume, which tends to move before anything else does. And keep a simple list of the third-party pages that mention you, revisited quarterly, because that list is the underlying asset and it either grows or it does not.
None of those are precise. Together they are enough to tell a real change from a quiet quarter, which is the only judgement you actually need to make.
None of this is fast. It is at least checkable, which is more than most advice in this area manages.
Related reading
- What a knowledge graph is. Why an engine consults one before it will risk naming anybody.
- AEO vs SEO. What actually changed when a ranked list became a composed answer, and what did not.
- The AEO glossary. Every term here defined, and marked by whether it comes from a standard or from marketing.
Shaban runs answer engine optimisation for B2B SaaS companies at Uncited Brands. He works the unglamorous end of the problem, mostly crawler access, retrieval diagnostics and measurement baselines, and publishes the tests that do not go his way alongside the ones that do.