What a frontlink is
A frontlink is the mention of a brand placed in the part of a publication that gets crawled most and reads most cleanly, carrying enough information for an automated system to reuse it without having to interpret it.

The short definition
Frontlink: a structured mention of an entity, placed in an accessible and crawlable source, written so that a language model can quote it verbatim in an answer.
Every word in that sentence carries a requirement. Structured mention means the identifying data sits next to the name. Accessible source means it is served as HTML and can be read without executing anything. Quotable means the sentence works outside its context, because that is how it reaches the answer: clipped.
Where the term comes from
The word existed before AI search and meant something more modest: a link placed on the front page of a publication rather than buried in an old article. It cost more because the front page carries the most internal authority and is the page a crawler visits most often.
That meaning has not been lost, it has been widened. The front page is still the most crawled place, and therefore still the best place to be. What changed is what being there is good for. It used to pass authority. Now it also puts you in the room at the moment a system gathers what it knows about a subject.
Frontlink is not academic vocabulary and no search engine has declared it a ranking factor. The established technical term for what appears in the answer is AI citation. Frontlink is a working name for the practice that produces those citations, and we use it because it names the difference in position against the backlink, and because the thing is already being bought and sold and needs a word.
What a frontlink is not
- Not a link bought in a blog network. A model does not count links, and a page nobody reads has no particular reason to end up in anyone’s set of sources.
- Not a press release pushed to a hundred identical portals. Text repeated word for word is collapsed into one source, not a hundred.
- Not a company listing with no substance. A name and a phone number give nobody enough to write a sentence about you.
- Not mass generated content. A model trained on synthetic text recognises the pattern, and the sites that publish it stop counting as sources.
The ten second test
There is a simple way to check whether a mention works as a frontlink. Copy it, strip away everything around it, and read it alone. If it still says who, what, where and since when, it is a frontlink. If pulling it out of the paragraph turns it into an empty phrase, it is not.
| Mention | Out of context | Works |
|---|---|---|
| “One of the best restaurants on the island.” | No name, no place, nobody saying it. | no |
| “Ca na Marga, Mallorcan market cooking in the old town of Palma, open since 2019.” | Name, category, place and date. Quotable as it stands. | yes |
| “Our client won an award this year.” | No brand, no award, no year. | no |
| “Talleres Sorell, structural carpentry in Manacor, 14 staff, third generation since 1968.” | Trade, municipality, size and continuity. | yes |
Why position matters this much
A system answering a question does not read the whole internet in that instant. It works with two materials: what it learned during training and what it retrieves in the moment, usually a small number of pages. In both cases the material that wins is the one that sits high, is recent, and reads without obstacles.
The front page of an active publication is crawled every few hours. An article from two years ago can go months without a visit. The same sentence, in those two places, is worth completely different amounts.
Does the frontlink replace SEO?
No. A site that cannot be crawled cannot be cited either. The usual technical work, speed, structure, markup and sitemap, remains the foundation. The frontlink sits on top of it.
How long does it take to show?
A new mention can appear in answers within days if it sits on a front page that is crawled daily. Consensus, which is what makes the result stable, takes months and several sources.
Can it be measured?
Yes, but not with rank tracking tools. You measure it by asking: fix a set of real questions, repeat them monthly across several systems, and record whether the brand appears and with what description.
Does it work in every sector?
It works hardest where people ask in natural language and expect a recommendation: hospitality, health, professional services, travel, property. It works least in catalogue purchases, where the buyer already knows the exact model they want.