Seer Interactive's 2026 analysis suggests LLM citation is partly post-hoc: the model selects brands from what it already knows, then retrieves sources to support the choice. If so, being known across many independent sources matters more than optimizing any single page. Off-site corroboration is the slowest and most durable part of AEO.
There is a pattern that anyone working in this area runs into within a few months. A site does everything right. Clean structure, answer capsules, valid schema, fast, crawlable. And it does not get cited, while a competitor with a mediocre page does.
The usual explanation is domain authority, which is a description rather than an explanation. The more interesting possibility is that the citation decision is not primarily about the page at all.
The post-hoc citation hypothesis
Seer Interactive's 2026 research suggests that LLMs may decide which brands to recommend from their training data first, then search for sources to support those recommendations.
If that is how it works, the mental model most of us carry is backwards. We imagine a system that researches a question neutrally and cites the best sources it finds. The alternative is a system that already has opinions, formed from everything it absorbed during training, and uses retrieval to justify them.
Why this is plausible
It matches how these models behave in other contexts. They are pattern-completion systems with strong priors. Asked to name good project management software, a model has an answer before it looks anything up, because the internet has been telling it for years.
It also explains the observation above. A well-built page from a brand the model has never encountered has no prior to attach to. A worse page from a brand mentioned in thousands of training documents does.
The honest caveat
This is one firm's analysis and it has not been independently replicated. It is likely to be partly rather than wholly true, and behaviour probably varies between systems and between query types. A factual question about a technical specification is likely retrieved more neutrally than a recommendation question about vendors.
I am presenting it because it is the most interesting finding in the field and because it is directionally consistent with what practitioners keep seeing. I would not build a strategy that only works if it is completely true.
What follows if it is even partly right
The practical implication is uncomfortable for anyone selling on-page work, including me.
On-page optimization determines whether you can be quoted cleanly. Off-site presence determines whether you are considered at all. The first is necessary. The second is decisive.
That reorders the work. It also explains why the businesses that appear in AI answers tend to be the ones that were already visible, and why a brand-new site with perfect structure sees so little return.
The Off-Site Corroboration Map
Five zones. They are not equally hard, and they are not equally valuable.
| Zone | What it is | Difficulty | Durability |
|---|---|---|---|
| 1. Structured listings | Directories, association and industry registries | Low | Moderate |
| 2. Profile consistency | Your presence on platforms you control elsewhere | Low | Moderate |
| 3. Editorial mentions | Someone else chose to write about you | High | High |
| 4. Discussion presence | Forums, communities, where practitioners argue | High | High |
| 5. Co-citation | Appearing alongside recognized names in the same context | Very high | Very high |
Zones 1 and 2 are a week of work and you should simply do them. Zones 3, 4, and 5 are the actual project.
Zone 1: structured listings
Get into the directories that matter for your category. For agencies, that means the recognized platforms buyers use to shortlist. For professional services, association and regulatory listings.
This is not link building in the old sense. The value is corroboration: multiple independent sources asserting that you exist, do this thing, and are located here. Consistency across them matters more than volume.
Zone 2: profile consistency
Every profile you control should describe you identically. Same name, same description, same specialism, all linking back to the site, with sameAs connecting them in your schema.
The failure here is not absence, it is drift. Profiles created over five years, describing five slightly different businesses. That ambiguity makes resolution harder for no upside.
Zone 3: editorial mentions
Someone independent writing about you, quoting you, or referencing your work.
For a small firm the realistic routes are narrow: contributing genuine expertise where journalists and writers are looking for sources, publishing something specific enough that others reference it, and being useful in public consistently enough that people know who to ask.
There is no shortcut and the paid versions are worthless, because a mention you bought sits on a site that sells mentions, which is a context models learn to discount.
Zone 4: discussion presence
The places where your topic gets argued about by people who do it. Industry forums, professional communities, specialist discussion venues.
This is the hardest zone and I suspect the most valuable, because it is the least gameable. It requires actually participating over time, usefully, without pitching. That is a slow individual effort rather than a marketing campaign, which is exactly why so few firms do it.
Zone 5: co-citation
Appearing in the same context as recognized names. A comparison article that lists you alongside established firms. A roundup that includes you. A discussion where you and known entities are mentioned together.
You cannot manufacture this directly. It emerges from the other four zones and from being genuinely worth including.
The small-site advantage nobody talks about
Here is the good news, and it is more significant than it first appears.
The 2026 research indicates that topical authority beats domain size for citation purposes. A small site covering one subject exclusively can outcompete a large general publication for that subject's citations, apparently because models treat focused sources as more reliable within their specialty.
Why this is unusual
In classic search, a small site fighting a large publisher generally loses. Domain authority is cumulative and largely a function of age and scale. There is no clever way around it.
In citation contests, narrowness is an asset. Being the site that only writes about one thing, thoroughly, appears to be a legitimate strategy rather than a limitation.
What to do with it
Pick a lane narrow enough that you can plausibly become the most thorough source in it, then go deeper than is comfortable. For most small firms that means one industry, or one problem, not one broad category.
The instinct to broaden, to cover everything a buyer might search, works against this. Breadth dilutes exactly the signal that gives a small site its only structural advantage here.
What not to do
Paid mentions and link schemes. Corroboration works because sources are independent. Buying it removes the property that made it valuable.
Mass PR distribution. A press release syndicated to two hundred sites is one source repeated, not two hundred sources agreeing.
AI-generated guest posts at volume. This is the current version of an old mistake, and it produces mentions on sites nobody reads, which is where models learn to discount.
Chasing every platform. Depth in two venues where your buyers actually are beats shallow presence in ten.
Treating this as a campaign. The most common failure is not doing the wrong activity, it is doing the right activity for eight weeks and stopping because nothing visible happened. Corroboration accumulates on a timescale that does not fit a campaign calendar, and a firm that participates consistently for two years will be in a position that no amount of concentrated spend can buy in two months. If you cannot commit to the slow version, it is more honest to skip this layer entirely and compete on the layers you can control, rather than starting and abandoning it.
A realistic twelve-month corroboration plan
The objection I get to everything above is that it sounds like a full-time job, and done badly it is. Here is what a sustainable version looks like for a small firm with no marketing department and a few hours a month.
The first month is the only front-loaded one. Audit where you currently appear by searching your business name and your founder's name and recording every independent result. Then close the obvious gaps in Zones 1 and 2: get listed in the three or four directories that genuinely matter for your category, and make every profile you already control say the same thing. Most firms discover in this exercise that they have four profiles describing three different businesses, which is free damage they have been carrying for years.
From month two, the work becomes a small recurring commitment rather than a project. I would budget roughly four hours a month, spent in one place rather than spread across five.
Pick the single venue where your buyers actually discuss their problems. For a firm serving dental practices that might be a practice-management forum. For one serving nonprofits it might be a specific professional association's community. The criterion is not audience size, it is whether the people there are the people who hire you. Then participate genuinely: answer questions in your area of expertise, without linking to yourself, for months, before you expect anything. The reason this works is the same reason it is rare, which is that it cannot be shortcut and most firms give up in week six.
Alongside that, publish one genuinely referenceable thing per quarter. Not a blog post restating consensus. Something with a number, a framework, or a position in it that another writer would need to attribute. Four of those a year is a realistic pace and it gives people a reason to mention you that does not depend on them liking you.
Then, twice a year, do a deliberate round of outreach to the places that cover your niche: trade publications, newsletters, podcasts serving your industry. Not a pitch about your firm, an offer of the specific expertise you have that their audience wants. Most of these go nowhere. One or two a year landing is a good outcome and compounds far beyond its apparent size.
Twelve months of that produces something like: consistent presence across the listings that matter, a recognizable name in one community, four referenceable assets, and perhaps two editorial mentions. That is a modest-sounding result and it is more corroboration than most small firms accumulate in five years of publishing blog posts nobody reads.
The discipline is choosing one venue and one cadence and not abandoning them in month three, which is when the absence of visible return is most discouraging and the work has barely begun.
Frequently asked questions
Do backlinks still matter for AI citation?
Links matter as one form of corroboration, but unlinked brand mentions appear to matter too, which is a meaningful difference from classic SEO. A model reading text that says your firm does X is receiving corroboration whether or not there is a hyperlink. That widens what counts as useful presence.
How long does off-site work take to show up?
Quarters, and possibly longer where training data is involved, since models are retrained periodically rather than continuously. This is the slowest layer in AEO and the reason to start it before you feel ready.
Can I do this without a PR budget?
Yes, and most small firms have to. The unpaid version is participation: answering questions publicly, publishing something specific enough to be referenced, and being consistently useful in a narrow venue. It costs time rather than money, which for a small firm is often the harder currency.
If citation is post-hoc, is on-page work pointless?
No. On-page work determines whether you can be quoted cleanly and attributed correctly once you are considered. Skipping it means being known and still not being used. The argument is about sequence and proportion, not about abandoning it.
How do I know if this is working?
Branded search volume is the earliest reliable signal. When off-site presence grows, people start searching your name, and that shows up in Search Console before citations do. Track it monthly.
The part of AEO you can control is the part that matters least, and the part that matters most is other people deciding you are worth mentioning. That is an unsatisfying conclusion and I think it is the correct one.
If you want help working out which of the five zones is your realistic next move, email adphconsulting@gmail.com. ADPH is a US-facing consultancy that delivers its work from the Philippines.