Long-tail keywords

A long-tail keyword is a query that occurs rarely on its own but belongs to a very large set of similarly rare queries, which together account for most of what people actually ask. The name comes from the shape of the frequency curve: a short head of high-volume phrases, and a long tail of specific ones that never repeats often enough to appear in a volume report.

The common shorthand — "three or more words" — is a proxy, not the definition. Length correlates with rarity because people add words to narrow a question, but the property that matters is frequency, not word count.

Why the tail is large

Because language admits far more ways to ask a question than there are questions. Every added constraint — a version number, a city, a comparison, an error message — multiplies the phrasings available. The result is measurable from inside a search engine, and Google has published the number: in its October 2019 announcement of BERT in Search, it stated that 15 percent of the queries it sees every day are ones it has not seen before. Novel queries arriving daily at that share is what a heavy tail looks like from the engine's side.

That single fact carries most of the practical consequence. A demand curve with that much novelty cannot be covered by targeting phrases someone has already recorded, because a large part of tomorrow's demand does not exist in any keyword tool today.

What changed when matching became interpretation

Tail strategy used to be lexical: find rare phrasings, publish a page for each, match the string. That worked while retrieval matched strings.

It stopped working when retrieval started interpreting them. The same Google announcement described the change as understanding queries where prepositions like "for" and "to" matter to the meaning, and said it was particularly useful for longer, more conversational queries. Semantic search generalised this: a query and a passage can match on meaning without sharing vocabulary. Two pages written for two phrasings of one question are now near-duplicates of each other competing for the same intent, which is a self-inflicted problem rather than coverage. See BERT for the mechanism and conversational search for the query style it enabled.

Why the tail is where assistants live

A rare, highly specific question is the exact case an assistant is good at: there is one right answer, it exists in some document, and stating it is more useful than listing ten links. That is why tail demand is the part of search most visibly absorbed by generated answers, and why the strategic question shifted from "which phrase do I target" to "is my passage the one that gets retrieved and repeated".

The eligibility path is unremarkable. Google's AI features guidance states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimizations necessary, and that to be used as a supporting link a page must be indexed and eligible to be shown with a snippet. So there is no tail-specific trick to buy. What there is, is a unit change: the thing selected is a section, not a page. See AI Overview.

Practically that means one question per section, a heading that states the question or the claim, and every qualifier written into the same sentence as the fact it qualifies. A section whose meaning depends on the two sections above it is correct on the page and wrong the moment it is quoted alone.

Failure modes

• Treating word count as the definition. Publishing longer phrases does not reach rarer demand if the phrases are common.
• One page per phrasing. Under semantic matching this produces siblings that cannibalise one another and read as templated.
• Only targeting recorded volume. With the share of never-before-seen queries Google reports, the tail you can look up is not the whole tail. Coverage of the underlying questions beats coverage of listed phrases.
• Burying the answer. A page that answers the question in its ninth paragraph, after a preamble, gives retrieval a chunk about the preamble.
• Assuming tail traffic converts because it is specific. Specificity indicates a precise question, not a purchase; the intent still has to be read.
• Chasing the phrase into the title only. A title that repeats a rare phrase over a body that does not answer it is the oldest version of this mistake and survives none of the above.

Frequently asked questions

Is it defined by length?

No. It is defined by rarity. Length is a correlate, because people add words to narrow a question.

How much demand is really in the tail?

Google reported in 2019 that 15 percent of daily queries were ones it had never seen before.

Does exact phrasing still matter?

Less. Retrieval interprets meaning, so several pages for several phrasings of one question compete with each other.

What is the unit to optimise now?

The self-contained section that answers one question, because that is what gets retrieved and repeated.