Keyword Volume vs Prompt Volume: What Do You Actually Build?

You prioritise it by building both numbers into the same table before you write a brief, not after. A topic with 20,000 monthly Google searches and almost no presence in AI conversations needs a different page to one with 5,000 searches and 250,000 prompts asking about the same problem. Treating them as the same research exercise is how good topics get buried in a spreadsheet nobody reopens.

This is for anyone still running keyword research the way it worked in 2022, one tool, one volume column, one ranked list. It will change how you decide what gets a brief this quarter and what doesn't.

Why isn't keyword volume enough to prioritise content anymore?

Keyword volume tells you how often people type a phrase into Google. It has never told you how often people ask the same question somewhere else, and increasingly they do. AI assistants absorb demand that never touches a search box, because a conversational question and a typed keyword are not the same behaviour, even when they are about the same problem.

I have sat with clients looking at a keyword report showing a topic at a few thousand searches a month, ready to deprioritise it, while the same topic comes up constantly when I manually test the same question in ChatGPT and Perplexity. The keyword tool isn't wrong. It's just measuring one room in a house where the conversation has spread to several.

This is not a small correction. Search Engine Land recently documented cases where a topic's prompt volume ran roughly fifty times its keyword volume, meaning a team relying on keyword data alone would never have put it on the roadmap at all. That is not a rounding error in a forecast. That is a topic that does not exist in the plan.

What is prompt research, and how is it different from keyword research?

Prompt research is the practice of measuring how often people ask AI assistants about a topic, in their own words, rather than measuring how often they type a short phrase into a search box. Keyword research answers "what do people type." Prompt research answers "what do people actually ask when they can phrase the question properly."

The difference in the queries themselves is the tell. A Google search for a service tends to run three to five words. The equivalent question to ChatGPT or Perplexity tends to run into full sentences, context included, because the person is describing a situation rather than guessing at a search term that will work. That is a wider net catching demand a keyword tool was never built to see.

Neither discipline replaces the other. Google search remains where most transactional and navigational demand still lives, and keyword data is still how you understand SERP competition and click potential. Prompt research adds a second lens onto the same underlying question: what does my audience actually want to know, regardless of which box they typed it into. This builds directly on the full keyword research process, it doesn't replace any part of it.

How do you actually compare the two for the same topic?

Put keyword volume and prompt volume in the same row of the same spreadsheet, for every topic you're evaluating, before you write a single brief. The comparison, not either number alone, is what tells you which kind of page to build.

For keyword volume, standard tools apply: Google Ads Keyword Planner as a base, cross-referenced against a paid tool like Ahrefs or Semrush for related terms and competition data. For prompt volume, purpose-built tools such as Profound's Prompt Volumes now model how often a topic appears across ChatGPT, Gemini, Claude and Perplexity conversations, alongside related phrasing. If that's outside your current budget, manual testing does most of the same job at smaller scale: ask the assistants the question yourself, in the way a real customer would phrase it, and note what comes back.

Once both numbers exist for a topic, four outcomes emerge, and each one calls for a different brief rather than a different tweak to the same brief.

Keyword-strong, prompt-weak. People search for this in Google. They aren't yet asking an assistant to walk them through it. Write the classic SEO page: keyword-matched title and H1, an answer early, and comprehensive coverage of what the current top results are missing. Structure it well regardless, because if AI search demand arrives for this topic later, there should be something worth citing.

Prompt-strong, keyword-weak. This is the bucket keyword research alone will never surface, and the one worth the most attention precisely because your competitors are still sorting their roadmap by keyword volume. Write for the answer, not the SERP: direct definitions, the actual question answered in the opening two sentences, structure built to be extracted and cited rather than scrolled through. This is where the distinct GEO layer on top of traditional keyword strategy actually earns its place in the brief.

Strong on both. Fund it properly. These are the pages worth building to rank in search and get cited in AI answers at the same time, which makes them the closest thing to a guaranteed return in the whole exercise.

Weak on both, or the prompt column comes back empty. Check the head term before writing the topic off. An empty result usually means the demand exists but is bundled into a broader category the tool tracks under a different name, not that nobody is asking. A narrow question can show nothing on its own while rolling up under a much bigger topic one level up.

Isn't this just another subscription and more busywork?

It's fair to ask whether this is worth adding a tool and a process step, particularly for a business already stretched thin on marketing time. The honest answer is that the tool is optional. The comparison is not.

You don't need Profound or any paid prompt-tracking product to do the core of this. Manually asking ChatGPT and Perplexity the ten or fifteen questions your best clients actually ask, in full sentences, and noting what kind of answer comes back and whether your brand or your competitors' show up in it, gets you most of the signal for free. What you cannot skip is the comparison itself. A keyword-only roadmap will consistently under-invest in exactly the topics where AI-driven demand is growing fastest, because the tool measuring your roadmap was never built to see that demand in the first place.

There's a connection worth drawing to Semrush's recent study of topic ownership in ChatGPT, which tracked over a thousand categories and found that ownership is decided at the topic level, across a cluster of related questions, not by winning any single prompt. Prompt research is how you find out which topics are worth building that cluster around before you commit the budget to build it.

How do you put this into practice without overhauling your entire content process?

Start with the topics you already suspect are underperforming their true demand, not with a full audit of everything you might ever write about. Pick the ten questions your sales conversations circle back to most often, the ones a prospect asks in slightly different words every time, and run each one through both a keyword tool and a manual AI test.

There's a version of this that any experienced practitioner develops without a tool at all, a feel for when a topic is bigger than its numbers suggest, built the same way a skateboarder learns to read a ledge before consciously calculating the angle. Enough reps against enough topics and you start noticing the mismatch before the spreadsheet confirms it. That instinct is worth trusting as a starting hypothesis. It is not worth trusting as the final answer, because the two-column comparison catches topics your instinct will miss precisely because they're quiet in the room you're used to listening in.

Once you've sorted a first batch into the four buckets, split your reporting the same way. Measure classic organic traffic and AI-referred traffic as two separate lines, not blended into one "organic" number, so you can tell whether a page built for the answer engine is actually earning attention there, rather than assuming it's working because overall numbers look fine.

Does this mean keyword research is becoming less important?

No. It means keyword research on its own has stopped being a complete picture of demand, in the same way that ranking well on Google stopped being a complete picture of visibility once AI assistants became a genuine discovery surface. The principle underneath both disciplines hasn't moved: understand what your audience actually wants to know, and build content that answers it properly. What's changed is how many rooms that question now gets asked in, and whether your research process is listening in more than one of them. This is exactly the kind of shift the Search Visibility Framework is built to map across all three layers at once.

Frequently asked questions