Why Search Volume Hides Your Best Content Opportunities

Most teams have overhauled how they write for AI engines, yet surprisingly few have rethought how they choose topics in the first place. That disconnect creates a real problem: content that is structured, summarized, and marked up for AI, but aimed at a keyword list built by sorting a spreadsheet on search volume. The writing has evolved, but the prioritization hasn’t.

Using search volume as a filter now actively works against you. It systematically excludes the highest-intent questions your buyers are asking, for one simple reason: nobody types those questions into a search box. They speak them to an assistant instead, and your keyword tools never see them. Here’s why that happens and what you should prioritize instead.

What Changed Under the Hood

Let’s look at the mechanics. When someone asks an AI assistant a question, it doesn’t run that question as a single search. Google’s documentation on AI features is explicit on this point: “Both AI Overviews and AI Mode may use a ‘query fan-out’ technique, issuing multiple related searches across subtopics and data sources to develop a response.”

Google also notes that its models identify further supporting pages while generating a response, which is why AI results tend to cite a wider spread of sources than a classic results page. ChatGPT does something comparable, rewriting the user’s prompt into search queries before retrieval. Although OpenAI uses a different name, the underlying idea is the same.

So searching hasn’t disappeared, it has moved. The model now does it on the user’s behalf, breaking one prompt into multiple queries and assembling an answer from passages across many pages. That single change breaks volume as a prioritization signal. Competition has shifted from ranking for one keyword’s SERP to being one of the sources pulled into a synthesis. In this era of search, your page can get cited for a sub-question you never targeted, or excluded from a topic you thought you owned because you only covered the headline term.

The Questions Your Tools Can’t Count

Keyword tools report demand for strings that people type. Prompts don’t look like that. They’re longer, conversational, and usually describe a situation rather than state a query. Here’s the same person, searching versus prompting:

| Search | Prompt | |——–|——–| | “best crm for small business” | “We’re a 12-person agency outgrowing spreadsheets. What CRM should we move to, and how painful is the switch?” | | “ecommerce seo audit” | “Our Shopify traffic dropped 30% since March. How do I work out what caused it?” | | “hreflang tags” | “We’re launching in France and Germany. Do we need separate sites, and how do we stop them competing with each other?” |

Look at the prompts. Each one carries constraints, a decision to make, and an implied objection. Each one is also worth far more to your business than the traditional search, because the person asking provides context and is closer to acting. You’ll also notice that each prompt returns nothing in a volume column, because hardly anyone types that exact sentence.

The trap is that if you sort by volume, traditional search wins every time, while the actual prompt never makes it onto the list at all. You end up prioritizing the queries that are easiest to count rather than the ones that are most valuable to answer.

4 Things to Prioritize Instead

These four shifts have made the biggest difference for me. They don’t require new tooling, just a change in what you’re looking for.

1. Prioritize the Sub-Questions, Not the Head Term

If a prompt fans out into sub-queries, then those sub-queries are your real targets. Take your head term and write out the eight to ten things someone would need answered before they could actually act on it. For “best CRM for small business,” that’s pricing tiers, migration effort, integrations, contract length, and what happens to your data if you leave. Then check whether your page genuinely answers those, or just gestures at them. Most pages do the latter.

2. Prioritize Entities and Concepts, Not Strings

Synthesis matches on meaning rather than exact phrasing, so exact-match repetition buys you far less than it used to. What earns you a place in the answer is covering the subject properly: naming the relevant products, standards, methods, and alternatives, and explaining how they relate to each other. In practice, that means fewer pages built around variants of the same phrase, and more pages built around a subject that’s covered thoroughly. Essentially, providing value.

3. Prioritize the Decision, Not the Definition

People don’t ask assistants for definitions much anymore, because they get those instantly. They use AI assistants to help them decide. So comparisons, selection criteria, trade-offs, and objections are now primary targets, rather than sections you bolt onto the end of a buyer’s guide. If your content can’t support a comparison, it won’t be much use to a model answering a comparison prompt.

4. Keep Volume Where It Still Decides

Search volume still has its advantages. Plenty of queries are still short, transactional, and settled by a normal SERP, such as brand terms, product terms, local intent, and “near me” searches. For those, volume remains a perfectly good signal, and you shouldn’t rebuild them around conversational prompts. Use both approaches and be clear about which of your pages belongs to which.

Where to Find The Raw Material

None of this needs a new tool. The inputs are already around you: People Also Ask and related searches, for the shape of the sub-questions. Reddit, Quora, and industry forums, where you’ll find contextualized conversations from real people. Your own sales calls and support tickets. And the assistants themselves. Ask one your head term and watch which sub-questions it chooses to answer. Sales calls are the underrated one here, because customers describe their problems to a salesperson in almost exactly the way they describe them to an assistant.

How to Rank a List With No Numbers on It

This seems counterintuitive to SEOs who have been in the industry for a while. If a question has no volume, how do you rank it against one that has 2,400 searches a month? You can’t, at least not on the same scale. What you can do is score on proxies:

• Business value. Does answering this move someone closer to buying? A question asked by 40 people mid-decision beats one asked by 4,000 people who are idly curious.I understand this is more nuanced and tedious than simply sorting a column, but this adaptation can give you lots more data and, hopefully, visibility leading to conversions.How You’d Know If It WorkedMeasurement and SEO reporting in this AI-driven era is where a lot of SEOs I speak to have dilemmas. However, you can boil it down to just two things to look at. First, whether AI sources are actually sending you people, and which pages they land on. That needs your analytics configured properly, because GA4’s default AI Assistant channel splits that traffic across several channels and undercounts it by default. Second, your ordinary organic performance. A page that covers its sub-questions properly tends to pick up long-tail rankings whether or not an assistant ever cites it.What you can’t do is attribute a specific citation back to a specific prioritization decision. Assistants don’t give you that data yet, so treat this as a directional program measured over time rather than a campaign that you can attribute precisely.This Isn’t an Obituary for Keyword ResearchI want to be very clear here: keyword research isn’t dead. What’s dead is treating a search volume column as the arbiter of what deserves a page. Working out what your audience wants to know, in what order, and with what intent behind it is more valuable now than it was when a SERP was the only surface you were optimizing for, because a model assembling an answer is far less forgiving of a page that half-covers a subject.Over to YouIf you want a concrete starting point: take your top 10 commercial pages and write out the prompt a real buyer would use to reach each one. Not the keyword, rather the whole sentence, constraints and all. List the sub-questions that prompt would fan out into, and mark the ones your page actually answers. Fix the gaps on your strongest existing page first, then go one-by-one in a linear order, sorted by commercial viability.Search volume was never the goal. It was a proxy for demand, and it worked well for as long as demand arrived as typed strings. Now that a growing share of it arrives as a described situation, the proxy has stopped tracking the thing it was standing in for. If your query lists still don’t sound like the way your customers talk, it’s worth sorting out before anything else.