The long tail is the set of rare, precise queries that each bring only a handful of visits a month, but that together account for a large share of search traffic. A bedding retailer will not rank easily for "mattress". It can answer "firm double mattress for lower back pain" and hundreds of neighbouring formulations, each marginal, all significant once added together.
The idea is old and it still works, provided you keep it up to date. Google still says a stable share of the searches it sees each day has never been seen before. Search Console deliberately hides part of those queries. And AI-generated summaries appear precisely on the long questions that make up the tail.
What follows is the definition, the figures that can be checked, and a method for finding these queries in your own data rather than in a keyword tool.
Where the phrase comes from
The term was popularised by Chris Anderson in an article for Wired, published in October 2004, then in the book that followed in 2006. His observation was about distribution: among online retailers, a few heavily demanded references form the head of the curve, while a long tail of rarely bought products adds up to a comparable volume of business.
Search marketing borrowed the metaphor and swapped products for queries. The head is generic keywords: "car insurance", "plumber", "CRM". The tail is the thousands of precise formulations people actually type.
Two clarifications matter. Length is not the criterion: low search volume combined with a more explicit intent is. A two-word query in a very specialised field belongs to the long tail; a six-word question asked constantly does not. And the boundary is continuous, with no official threshold separating head from tail.
What Google says about unseen queries
When Google announced that BERT had been integrated into search, on 25 October 2019, Pandu Nayak, then vice-president of Search, wrote on the official Google blog that 15% of the queries seen each day had never been encountered before. This is Google describing its own worldwide search volume, with no published methodology: treat it as an order of magnitude from the vendor, not as independent research.
The figure has not been revised since. In March 2025, at Search Central Live in New York, John Mueller said the proportion was still around 15% and that he was surprised by it, the arrival of language models having failed to move the indicator (reported by Search Engine Journal, March 2025).
Tool vendors point in the same direction. Ahrefs reports that keywords recording fewer than ten searches a month make up close to 93% of its US keyword database (Ahrefs, article updated in May 2026). Scope matters: this is an SEO tool vendor's database, for the US market, with volumes estimated rather than measured by Google. It indicates dispersion, not real traffic.
The consequence is simple: no keyword list will ever cover the tail. You do not target it query by query; you create the conditions that let you capture it.
Head, middle and tail: practical markers
The table below summarises the differences usually observed between the three zones of the curve. These are working markers, not thresholds.
| Criterion | Head (short tail) | Middle (mid tail) | Long tail |
|---|---|---|---|
| Example | boots | women's walking boots | waterproof wide fit walking boots for women |
| Typical length | 1 to 2 words | 2 to 4 words | 4 words and up, often a question |
| Indicative monthly volume | High | Medium | Low to very low, sometimes never seen before |
| Intent | Ambiguous, needs qualifying | Partly qualified | Explicit: comparison, purchase, a specific problem |
| Competition | Strong, established sites | Variable | Light, sometimes no dedicated page at all |
| Effort to rank | High: authority, links, time | Medium: solid content and internal linking | Low: one well-written section often does it |
| What you do with it | Pillar page, visibility objective | Category or service pages | Sections, FAQs, detailed specifications, practical articles |
Finding your long-tail queries in Search Console
Volume estimation tools ignore what is rare, by construction. Search Console tells you what Google actually associates with your pages. In an audit this is the first place to look: queries that collect impressions without clicks show where an existing page can be completed, which costs far less than new content. The same data is the natural starting point when you plan an SEO audit.
Filter with regular expressions
In the Performance report, under the Queries tab, the filter offers a "Custom (regex)" option accepting RE2 syntax (Search Console documentation). Two patterns are enough to start.
^(how|why|what|which|when|can|is|are|does|should)
(price|cost|pricing|quote|quotation)The first isolates questions. The second isolates a semantic field by listing the variants in brackets, here to separate commercial intent from informational intent. Then sort by impressions and look at the average position. Queries sitting between eighth and twentieth place, with impressions and few clicks, are usually the most profitable: Google already knows the page, it is only missing an explicit answer.
Limits worth knowing
Search Console does not show everything, and this is documented. Google explains that anonymised queries, meaning those not issued by more than a few dozen users over a two to three month period, are not displayed, to protect searchers' privacy. They are counted in the chart totals but appear in no row of the table (Google Search Central, October 2022). The finest part of your long tail is therefore invisible by design, and the gap between the chart's total clicks and the sum of the table rows roughly measures it.
The second limit comes from the same source: the export from the interface is capped at 1,000 rows, while the Search Analytics API and the Looker Studio connector go up to 50,000 rows per day, per site and per search type. Above a few thousand queries a month, a manual export gives you a truncated picture.
One prerequisite is easy to forget: a page only appears in these reports once it is crawled and indexed. If whole sections are missing, the problem is not the tail but how your pages get indexed, and that comes first.
What AI summaries change for these queries
AI Overviews appear mostly where the long tail lives. A study by the Pew Research Center, published in July 2025, followed 900 US adults who volunteered from the KnowledgePanel and analysed 68,879 Google searches from March 2025, of which 12,593 included an AI-generated summary. Three results matter here.
- 53% of queries of ten words or more triggered an AI summary, against 8% of one or two word queries.
- 60% of queries starting with a question word (who, what, when, why) triggered a summary.
- On results pages carrying a summary, participants clicked a standard result link in 8% of visits, against 15% on pages without a summary; links inside the summary were clicked in 1% of visits.
Keep the scope in mind: the United States, one month of collection, 900 people who agreed to have their browsing tracked. These figures do not transfer mechanically to the United Kingdom or to your sector, but the direction is clear. On long, purely informational questions, impressions can rise while clicks stay flat.
The conclusion is not to abandon the long tail, but to sort it. A tail query with commercial, local or transactional intent still holds real click potential: "quote", "near me", "versus", a product reference. Those are also where local search visibility does most of its work, because the answer people want is a business they can reach. A generic three-line definition holds much less.
Working the tail without spawning thin pages
The classic mistake is to create one page per query variant. Google treats these as low-value doorway pages, and internal linking gets diluted. A few safer principles apply.
- Group by intent, not by keyword. If ten formulations call for the same answer, they belong on the same page, with subheadings that use the real wording.
- Enrich existing pages first. An indexed page that gains a "which size to choose" section captures the matching tail faster than a new page does, and it is the cheapest form of SEO content work.
- Answer before you elaborate. A direct two-sentence answer under the heading, then the detail. That is the format search engines lift, featured snippets included.
- Use your customers' vocabulary. Questions asked on the phone, in quote requests and in after-sales support beat any keyword generator.
- Measure by group, not by query. Track clicks across a set of pages over three months. At the scale of one tail query, statistical noise dominates.
The long tail remains one of the few areas where a small business can get ahead of an established player: competition is light and the expected answer is precise. What it demands is consistency, with pages completed regularly from what Search Console reports, rather than a one-off campaign.
Common questions
How many words does a long-tail query contain?
There is no official threshold. Three words or more is common, but the real criterion is low search volume combined with a precise intent. A short, highly specialised query belongs to the long tail too.
Is the long tail still worth the effort now that AI summaries exist?
Yes, but it needs sorting. According to the Pew Research Center (July 2025, US panel), 53% of queries of ten words or more triggered an AI summary, and the click rate to a standard result was lower on those pages. Tail queries with commercial, local or transactional intent remain the most useful to work on.
How do I find long-tail keywords without a paid tool?
Search Console is enough to start. The Performance report lists the queries actually associated with your pages, with a regex filter to isolate questions. Google's autocomplete suggestions and the questions your customers ask fill in the rest.
Should I create a page for every long-tail query?
No. Formulations that call for the same answer should live on a single page, structured with subheadings. Multiplying near-identical pages produces low-value content and dilutes internal linking.
Why does Search Console not show all of my queries?
Google does not display what it calls anonymised queries, those not issued by more than a few dozen users over two to three months, to protect privacy. They are included in the chart totals but absent from the table, which explains the gap between the two.
How long before long-tail work shows results?
It depends on the state of the site, but enriching a page that is already indexed generally produces signals sooner than publishing a new one. Measure by group of pages over at least three months; at the scale of one query, the variations cannot be interpreted.