Updated
Traffic to sites is down because people are reading more inside the LLMs, Google included. The order of work has changed with it, get the site right first and the schema right, then write content that registers in an LLM and gives the reader a reason to click through. Long-tail is where that click still lives.
Long-tail keywords are specific, multi-word search queries that each carry low volume but together make up a large share of what people type into a search box. “SEO” is a head term. “How to improve SEO for a new online store selling handmade jewelry” is a long-tail query. Long-tail has long been the strategy for smaller publishers who cannot compete on head terms, and in 2026 it is also the strategy for surviving AI search disruption, because the broad queries are the ones being answered before anyone reaches your site.
Where the head-term clicks went
Google kept the head-term clicks for itself. Pew Research Center tracked the browsing of 900 U.S. adults in March 2025 and found that when Google showed an AI summary, people clicked a traditional result on 8 percent of visits, against 15 percent when no summary appeared, and clicked a link inside the summary on 1 percent of visits. A summary shows up first on the broad, definitional query, so the loss falls hardest on the pages that used to win “what is SEO” and lightest on the pages that answer a question only a practitioner would type. A traffic drop that clusters on your broadest pages fits that pattern.
Why long-tail survives AI answers
AI search tools are most effective at answering broad, common questions, so “What is SEO?” gets a full AI answer and the reader has no reason to click through. “How do I recover organic traffic after replatforming a store and changing the product URLs?” is specific enough that the tools need to find and cite specialized content.
The more specific the query, the more the user needs detailed, experience-based content, the kind that comes from having done the work, and the answer has to be fetched from a source. That sourcing gap is the opening for your pages. Picking which queries deserve a page comes first, and once an engine arrives, the companion piece on what gets your content cited in AI search covers what earns the citation.
What Content Cucumber measured on its own site
Content Cucumber runs a pulse on contentcucumber.com twice a month, asking four answer engines the broad category question, who is a content marketing and writing agency for ecommerce brands, and scoring how the site shows up. The September 1, 2026 run put the blended AI visibility score at 55 out of 100, with OpenAI at 59, Perplexity at 58, Kimi at 53 and Gemini at 50, and share of voice on that category question at 0.0 percent on each of the four engines, a figure that has read zero since the first run in July. Content Cucumber, a company that writes about search for a living, does not get named when an engine is asked the category question, and the answer gets composed without sending a click to anyone. I am not conceding that question. Content Cucumber is passionate about the category, humans are back, and people now recognize AI-created content when they read it. In Search Console the queries producing impressions for the site are the specific ones, a phrase like “enterprise search engine optimization software” sits at position 21 while the one-word head terms send nothing. The commerce LLM discovery work starts from that split.
Which queries still send traffic
You still get the click where you are the cheapest source of the answer, and the Search Console exports behind Content Cucumber’s audits keep turning up the same shapes.
The vendor-hunting query comes first. Someone typing “enterprise search engine optimization software” wants a shortlist and a demo, and a summary cannot book the demo, so the engine hands off. The situation query is the second, “Organic traffic dropped after a replatform and the product URLs changed” describes a mess only a person who has cleaned one up can answer.
- The comparison with a constraint. Two named options plus a budget, a catalog size or a team size.
- The question that needs your data. Your return rate, your lead time, your pricing, which no engine has unless you published it.
Head terms, definitions and “best X” lists without a constraint are the queries an engine now answers in place. I would not pick one shape and bet on it. A merchant has to try all four, because the engines are changing what they hand off week to week, so test the pages against Search Console and adjust weekly.
How to pull your real long-tail list from Search Console
The list tied to your own site is already sitting in Search Console, so start there. Then add your customer questions, since the specific questions your audience asks are natural long-tail targets. “How do I write product descriptions for technical B2B products?” is a question people type, with search demand behind it. Google’s “People Also Ask” and “Related Searches” sections list the questions around a head term, AnswerThePublic and AlsoAsked.com map them across a topic. A merchant needs those tools alongside the Console export, because the Console shows only the queries where you already appear.
Open Performance, set the range to the last three months, export the queries, then filter to rows with impressions and a click-through rate under 2 percent at positions 8 through 30. Keep the rows phrased as questions or situations and drop the one and two word head terms. What remains is the list of queries an engine is still handing off, where you are already listed without yet being the page it picks. The export is a short job, and the keyword research primer covers the rest of the mechanics.
Writing the page once you have the query
Each long-tail piece needs to go deep on its one scenario and answer that question in full, and that depth is what makes the page useful to readers and to answer engines looking for a source on a specific topic. Organize related long-tail content into clusters that link to a broader pillar page, which builds topical authority while serving specific queries. The same clustering held for spoken queries when voice search optimization was the frontier, and the answer engines inherited that job.
Where long-tail stops working
A long-tail page does not rescue a head-term page. If “what is SEO” carried your traffic, the summary owns that query now, and specific pages will not buy it back. Long-tail also stops paying when the specific question has no buyer behind it, since a query with impressions and no commercial intent brings a reader who is not shopping. Pick the questions your customers ask when they are close to buying. If the answer to that is to write all of them because long-tail pages are cheap, Content Cucumber works with the merchant on a workflow that makes financial sense at that volume and keeps a human over every post before it goes out.
Content Cucumber’s AEO and SEO audit pulls the query-level answer out of your own Search Console, shows which queries went with the traffic you lost and measures where the answer engines name you. The SEO content service writes the pages for the queries that still send a click.
Explore our writing services
Content Services →Frequently Asked Questions
Pew Research Center tracked 900 U.S. adults in March 2025 and found that when Google showed an AI summary, people clicked a traditional result on 8 percent of visits, against 15 percent when no summary appeared. Summaries show up first on broad, definitional queries, so the pages that used to win head terms lose the most traffic.
The queries where a site is still the cheapest source of the answer. Four shapes hold up in Search Console data, the vendor-hunting query, the situation query that describes a specific mess, the comparison with a constraint such as a budget or catalog size, and the question that needs the site's own data. Head terms, definitions and unconstrained best-of lists get answered in place by the engine.
The September 1, 2026 pulse asked OpenAI, Perplexity, Kimi and Gemini the broad category question for a content marketing and writing agency for ecommerce brands. The blended score was 55 out of 100, with OpenAI at 59, Perplexity at 58, Kimi at 53 and Gemini at 50, and share of voice on the category question was 0.0 percent on each engine. In Search Console the queries producing impressions are specific ones such as enterprise search engine optimization software.
Open the Performance report, set the range to the last three months, export the queries, then filter to rows with impressions and a click-through rate under 2 percent at positions 8 through 30. Keep the rows phrased as questions or situations and drop the one and two word head terms. The export takes about twenty minutes.
A long-tail page does not rescue a head-term page, since no number of specific pages buys back a query the AI summary now owns. Long-tail also stops paying when the specific question has no buyer behind it, so the pages to write are the questions customers ask right before they buy.





Really interesting take on how AI search is changing the way SEO traffic works. I especially liked the focus on long-tail queries and specific search intent rather than relying only on broad keywords. The Search Console approach also gives a practical way to identify opportunities that can still generate meaningful clicks.