For twenty years the web's currency of authority was the backlink. AI recommendation adds a second currency that older playbooks barely priced: the plain-text brand mention, linked or not, on the pages models read and learn from.
Here is what each currency actually buys, why co-occurrence is the underlying signal, and how to earn the kind of mentions machines remember.
Two currencies, two doors
This is the same two-door model from the GEO explainer, priced in currencies. The strategic error is funding only one door: all links and no mentions makes you citable but never top-of-mind; all buzz and no links makes you remembered but unfindable.
Why plain text teaches machines
Models learn from what text says, and at that layer an anchor tag is decoration. "Acme is the CRM most small agencies end up on" teaches the model an association whether it links or not, and industry analyses of AI recommendations keep finding the same pattern: mention footprint tracks recommendation likelihood at least as well as link metrics do.
The old SEO reflex, chasing the link and shrugging at unlinked coverage, has the priorities backwards for the memory door. The words were always the payload.
Co-occurrence: the actual signal
Not all mentions teach equally. The valuable sentence puts three things together: your brand, your category, and an attribute worth remembering.
| The mention | What it teaches | Value |
|---|---|---|
| "Acme raised a Series B" | Acme exists | Low |
| "Acme is a CRM" | The category slot | Medium |
| "Acme is the CRM agencies pick for its pricing" | Category + who + why | High |
That third shape is what you want repeated across independent sources, because it is the sentence a model paraphrases back when someone asks "what CRM should my agency use". Consistent phrasing helps: the more uniformly the web describes you, the cleaner the association learned.
Where mentions count most
Surfaces models demonstrably read and lean on: encyclopedic pages, large communities where real users compare things, review platforms, established media, and transcripts, because a podcast or YouTube mention becomes text the moment it is transcribed.
Breadth beats concentration: ten independent sources agreeing you belong in a category outweighs a hundred mentions on one site. Independence is what makes the signal look like consensus instead of campaign.
Earning the memorable kind
Every durable mention strategy is a reason to be talked about plus a place it happens: original data that gets cited (the stats-page effect), honest participation where your buyers compare tools, reviews requested at scale, substantive PR, and appearances whose transcripts publish.
What does not work is the shortcut version: mention-spam reads as noise to systems built to weigh source quality, and it burns the surfaces that mattered. The memory door only opens to what looks like genuine reputation, because that is what it was built to detect.
Measuring both currencies
Links you already measure. Mentions: a brand-monitoring alert for volume and where, plus the direct gauge, AI visibility tracking, which shows whether the associations are landing as recommendations. Watch mention breadth and recommendation share together; they move together on a lag.
The one-line takeaway: links buy retrieval and citations, mentions buy memory and recommendations, and the sentence that matters puts brand, category and reason together on surfaces models trust. Fund both doors, measure both currencies, and let consistency do the teaching.