Why your long searches are worth more than they look
Our keyword tool says nobody searches for these long questions. Why would we spend time on them?
Because your keyword tool is guessing and your Search Console is not. A keyword tool estimates demand across the whole market from a sample. Search Console shows the searches where Google actually displayed your site, with impressions, clicks and an average position. When a four or five word question shows up there, a real person typed it and Google already thought your site was relevant enough to show.
We see this on our own site. Over 90 days to 16 October 2026, dlvx.co earned impressions for full questions like "does claude work in thailand" (10 impressions, average position 11.3) and "is claude available in thailand" (3 impressions, position 10). The second phrasing is below our 10 impression filter on its own; we group near duplicates first, and the filter applies to the group. No keyword tool would list either as worth targeting. They are exactly what an owner asks before buying help with AI in Thailand.
That first number matters. Search Console hides rare queries for privacy, so a big share of your clicks arrive from searches you will never see by name. The long queries you can see are the visible edge of a much larger pile. Every one you answer well tends to pull in neighbours you cannot see.
Old studies said the same about volume. An Ahrefs experiment from October 2022, on its own blog, found that keywords estimated at 10 searches or fewer a month did deliver about that many impressions, with fewer than 1 percent passing 100 (Ahrefs, 21 October 2022). It is a small, single site test and four years old, so we use it only for the direction: small estimates are not zero.
Its AI optimization guide says "you don't have to worry that you don't have enough 'long-tail' keywords or haven't captured every variation of how someone might seek content like yours." We agree with the point, and this method is not keyword coverage. We only act on questions your own Search Console already shows with impressions, and we answer them on the page that ranks, so the page says plainly what people are already asking it. The effect of each change is measured, not assumed.
Are long queries really AI searches? We do not know, and it does not matter
You will hear that the long queries in Search Console are AI fan-out searches leaking through. It is a nice story. As of October 2026 nobody has shown it. Google does say that traffic from AI Overviews and AI Mode is counted in the normal Search Console performance report under Web (Google, AI features and your website), and AI fan-out queries are long and specific, so some overlap is plausible. Plausible is not proven.
Claim: long Search Console queries are AI fan-out. Grade: claimed, unproven. What is proven: they are real impressions on your own property, with a position you can move. That alone is reason enough to work them, so we never need the AI story to justify the effort, and we never sell it to a client.
Holding that line keeps the reporting clean. We report long-tail wins as search wins: positions and clicks for named queries. If an AI visibility number moves at the same time, that is measured separately, with its own method. See measuring AI visibility.
How to pull them out of Search Console
You can do this in the Search Console interface with a custom regex filter, or through the API. Our engine runs it monthly through the API. The steps are the same.
- Pick the window. The last 90 days, ending 3 days ago. Recent days are still filling in, and 90 days gives enough impressions on rare queries to see a pattern.
- Filter to queries of four or more words. In Search Console, add a query filter, choose "Custom (regex)", and paste the expression below. It keeps queries with at least four words.
- Filter by buying country. One country at a time, for each market you sell in. A query from a country you do not serve is a distraction, however big it looks.
- Add the page dimension. You need to know which of your pages Google showed for each query. In the interface, click a query and switch to the Pages tab. Through the API, request query and page together.
- Remove brand queries. Search Console now has a branded queries filter, announced on 20 November 2025 and available to all eligible sites from 11 March 2026 (Google Search Central). People searching your name already found you.
- Keep the ones worth a look. At least 10 impressions in the window, and an average position worse than 1.5. Position 1 is already won.
^([^ ]+ ){3,}[^ ]+$Read the expression as: three or more groups of "some characters then a space", then one more word. That is four words or more. Thai text does not separate words with spaces, so for Thai queries we review the list by hand rather than trusting the count.
If your shop runs on a different subdomain inside the same Search Console property, its queries land in the same list. We separate pages on other hosts before scoring, otherwise a blog fix gets credited for shop traffic.
How we decide which ones matter
A raw list of long queries is a few hundred lines of noise. We score each one so the top five are the ones where a small edit could move real clicks.
| Factor | How it counts | Why |
|---|---|---|
| Impressions | The base of the score | More people saw you, more people could click. |
| Click gap | Impressions multiplied by the gap between the click rate at position 1 and the click rate at your current position | A query at position 9 has more to gain than one at position 2. |
| Positions 4 to 15 | Weighted up, times 1.5 | Close enough that one good section can move it onto page one or up the page. |
| Beyond position 30 | Weighted down, times 0.3 | A section alone rarely lifts a page from that far back. |
| Matches a harvested fan-out query | Weighted up, times 1.3, when half the words or more overlap | Two independent signals agree on the same wording. |
The click rate curve we use is our own internal heuristic, built from typical rates by position. It is not a measured curve for your site, and click rates have moved a lot since AI Overviews arrived. Ahrefs measured position 1 click rates on AI Overview keywords falling from 7.3 to 1.6 percent between December 2023 and December 2025, on 300,000 keywords, desktop only (Ahrefs, 4 February 2026). That is why we use the curve to rank queries against each other, never to forecast traffic.
Then we group near duplicates. "office coffee machine rental bangkok" and "rent coffee machine office bangkok" are one intent. Two queries that share most of their words, or share many words and point at the same page, become one row with both phrasings listed.
So the output is a list of keywords to rank for?
No. The output is at most five intents per month, each with the page it belongs on, the exact buyer phrases, and one decision: add a section, or build a page. Five is deliberate. A small site that ships five good sections a month beats one that plans fifty.
Add a section or build a page?
Writing pages is cheap now. Why not make one page for each of these questions?
Because Google is explicit that it is a bad idea. Its July 2026 guide warns against separate content for every query variation, and its spam policies treat many pages made mainly to manipulate rankings as scaled content abuse, however they are produced (Google AI optimization guide, updated 10 July 2026; spam policies, updated 28 August 2026). Cheap to write is not cheap to own: every thin page dilutes the site and needs links, indexing and upkeep.
| What Search Console shows | Decision | Why |
|---|---|---|
| A dedicated page already ranks for the query | Add a section to that page | Google already connects the query to the page. Answer it properly there. |
| Your sitemap has a page whose topic clearly matches, even if it is not the one ranking | Add a section to that page, and link to it | Point the signal at the right page instead of creating a rival. |
| Only your homepage or a language root ranks, and no page matches | Consider one new page | There is a real intent with nowhere to live. One page, after a duplicate check. |
| The answer needs a claim you cannot back up | Decline the query | A section that invents a sourcing, safety or medical claim costs more than it earns. |
In our runs, most long-tail intents end up as one new H2 and one paragraph on a page that already exists. That is the cheapest change, and it strengthens the page that already has Google's attention.
How to write the section
- The heading is the buyer's question. In the page's language, close to how they typed it, made grammatical. "Does Claude work in Thailand?" not "Claude Thailand availability information".
- The first sentence answers it. Yes, no, it depends on one named thing, or a number. The reader and the engine should both have the answer before the second sentence.
- Then the useful detail. Two to four short paragraphs, a list or a small table. Steps, conditions, prices only if they are already published.
- Only facts you already hold. Use what is on the site, in your own records, or what the owner has stated. No new prices, product names, guarantees or health effects.
- Link out from it. If the answer leads to a service, link to that service page with a descriptive anchor.
Before and after, using an illustrative service page:
| Heading | First sentence | |
|---|---|---|
| Before | Our flexible options | We offer a range of flexible solutions tailored to your needs. |
| After | Can we rent an office coffee machine month to month? | Yes. The rental runs month to month after the first three months, and service visits are included. |
Google does not prescribe a length or a structure for this, and it says there is no need to chop pages into small pieces for AI features (Google, July 2026). Answer first is simply better writing. It helps a skimming human, a snippet and an AI answer equally.
Some pages, like product templates, have no slot for an extra section, or sit in files we cannot change. Then we put the section on the closest guide we can edit and link from it to the page that ranks, using the buyer's phrase as the anchor.
A worked example from our own site
Our own Search Console gives a clean case. In the 90 days to 16 October 2026, two long questions showed up together: "does claude work in thailand" and "is claude available in thailand". They are one intent: a business owner checking whether Anthropic's assistant can be used from Thailand before investing time in it.
| Step | What we found | Decision |
|---|---|---|
| Intent | Two phrasings, same question, both around position 10 to 11 | One row, two phrasings |
| Page | We already publish a Claude in Thailand page about availability and pricing | Work on that page, no new page |
| Heading | Make sure "Does Claude work in Thailand?" exists as a heading in the buyer's words | Gap A: wording |
| First sentence | A direct yes or no with the condition that matters, from facts already on the page | No new claims |
| Links | From our Claude model guide to the page, anchor "Claude in Thailand" | Two to five links |
| Check | Position and clicks for both phrasings, 4 weeks after shipping | Logged as a decision |
This is how the method reads it, step by step. Positions around 10 are the sweet spot in our scoring. Page one is in reach, the page is already relevant, and the edit is one heading and a paragraph. You can see the live page at dlvx.co/th/claude-in-thailand and the related Claude model guide.
What is hype, and what is a waste of time
| Idea | Verdict | Why |
|---|---|---|
| A page per long-tail question | Against the rules | Scaled content abuse under Google's spam policies and its July 2026 guide. |
| "Long queries are AI searches, so this is AI optimization" | Hype | Unproven as of October 2026. Real impressions are reason enough. |
| Trusting a keyword tool's zero | Skip it | Your own Search Console is the better source for your own site. |
| Answering in a new FAQ block for rich results | Skip it | FAQ rich results stopped showing on 7 May 2026. A visible question and answer still helps readers. |
| One new heading and a direct answer on the ranking page | Works | Cheap, reversible, measured per query. |
| Chasing queries from countries you do not serve | Skip it | Impressions without buyers. |
Measuring it, and a checklist
We check each shipped section 4 weeks after it goes live: average position and clicks for each phrasing in that row, on that page, in the buying country. We compare with pages we did not touch in the same language over the same weeks, because the whole site moves with the season and with Google.
Google updates are the other confounder. The Search Status Dashboard listed a March 2026 core update starting 27 March, a May 2026 core update starting 21 May, and a September 2026 spam update that started on 24 September and ran 13 days and 16 hours (Google Search Status Dashboard). If a check window spans one of these, we mark the result inconclusive unless the change is clearly larger than normal noise. Google also says improvements can take months to register after a core update (Google, core updates and your website).
Checklist
- Pull. 90 days ending 3 days ago, regex for 4 or more words, per buying country, query and page.
- Clean. Drop brand queries, other hosts and anything under 10 impressions or already at position 1.
- Score and group. Click gap, weight positions 4 to 15, group near duplicates into intents.
- Decide. Section on the ranking page by default. New page only when nothing matches.
- Write. Buyer's question as heading, answer first, facts you already hold.
- Check. Position and clicks per phrasing after 4 weeks, against untouched pages, with update dates in mind.
This is step seven of the Get Found track. It pairs with query fan-out, which gives you the AI side of the same vocabulary, and with internal linking, which gets the improved page the attention it needs.
Sources and dates
Every external fact in this guide was read on 20 October 2026. Each source shows its publisher and the date it was published or last updated, and we flag what changed in 2026. Studies are their authors' measurements, usually correlations, not promises of results. Our own numbers come from our own Search Console and tools and say so where they appear.
- AI Features and Your WebsiteGoogle Search Central · Last updated 10 December 2025 · Read 20 October 2026.
- Google's Guide to Optimizing for Generative AI Features on Google SearchGoogle Search Central · Last updated 10 July 2026 · Read 20 October 2026. Changed in 2026: Guide added 2026-05-15 (changelog); llms.txt clarification note added 2026-06-15; page last updated 2026-07-10.
- Introducing the branded queries filter in Search ConsoleGoogle Search Central · Published 20 November 2025 · Read 20 October 2026. Changed in 2026: Page update dated 2026-03-11: filter now available to all eligible sites.
- Latest Google Search Documentation Updates (FAQ rich result entries, May and June 2026)Google Search Central · Undated page · Read 20 October 2026. Changed in 2026: No standalone blog post found; announced only in the documentation changelog.
- Google Search's core updates and your websiteGoogle Search Central · Last updated 10 December 2025 · Read 20 October 2026. Changed in 2026: Changelog 2025-12-09 added a section on smaller, unannounced core updates.
- Spam policies for Google web searchGoogle Search Central · Last updated 28 August 2026 · Read 20 October 2026. Changed in 2026: Site reputation policy section updated 2026-08-28 for the EEA; 2026-04-13 back button hijacking added to malicious practices; 2026-05-15 clarification that spam policies apply to generative AI responses.
- Google Search Status Dashboard: ranking incidentsGoogle Search Status Dashboard · Undated page · Read 20 October 2026. Changed in 2026: Listing shown on read; older 2025 incidents may exist beyond the summary view.
- Growth Engine scoring, indexing and probe rulesDLVX (Growth Engine operating notes) · Our own measurement, data to 19 October 2026 · Read 20 October 2026.
- dlvx.co Search Console performance, 19 July to 16 October 2026DLVX (own Search Console, dlvx.co) · Our own measurement, data to 16 October 2026 · Read 20 October 2026.
- Update: AI Overviews Reduce Clicks by 58%Ahrefs · Published 4 February 2026 · Read 20 October 2026.
- Should You Target Zero-Volume Keywords? It DependsAhrefs · Published 21 October 2022 · Read 20 October 2026.
- Initial Research: Gemini 3 Query Fan-OutsSeer Interactive · Published 21 November 2025 · Read 20 October 2026.
Want this run on your own site? Start with the AI visibility diagnostic, or see the whole Get Found track.