Keyword research for SaaS when the tools say zero volume

The first keyword list I built for a vertical SaaS client had 340 terms and I was pleased with it. Six months later, the four pages that produced every single demo request from organic came from keywords that were not on it. All four came from a sales call transcript.

That was 2021 and it permanently changed how I do this.

Why the tools fail in narrow categories

Keyword volume estimates come from clickstream panels — samples of real browsing behaviour, extrapolated to the whole population. The extrapolation works when the sample contains enough people doing the thing. For “running shoes” it does. For “HIPAA compliant scheduling software for dental practices” the panel might contain nobody at all, so the tool reports zero, or 10, or some number generated with enormous error bars it does not show you.

You can test this yourself in about twenty minutes. Open Search Console, filter to queries with impressions but no clicks, and cross-check a handful against your keyword tool. On the last three accounts I did this for, between 30% and 45% of queries with more than 50 monthly impressions were reported as zero volume by the tool.

Those are real searches by real people, measured by Google rather than estimated by a vendor. And nobody is competing for them, because everyone else also sorted by volume and filtered out the zeroes.

Sales calls are the highest-yield source, by a distance

Get access to Gong, Chorus, or whatever your team records with. Listen to twenty calls from the last quarter, weighted toward closed-won and closed-lost. You are listening for one thing: how the prospect describes their problem before your AE reframes it in your product’s language.

The gap between those two vocabularies is where your missing keywords are. Your team says “revenue intelligence platform.” The prospect says “something that tells me which deals are actually going to close.” Guess which one gets typed into Google.

Transcript search makes this faster than it sounds. Search for phrases like “we were looking for,” “we tried,” “the problem was,” “does it,” and “can it.” The last two surface feature questions, which map almost one to one onto long-tail commercial queries.

On a compliance client, one AE call contained the phrase “we needed something that would keep the auditors off our back without hiring a compliance person.” That became a page targeting a cluster around outsourced compliance for small teams. It brings 70 sessions a month and has converted at 8.4%. Ahrefs still says the head term has zero volume.

Mining Search Console properly

Most people open the Performance report, sort by clicks, and close it. The useful data is elsewhere.

Filter to queries with impressions above 50 and average position between 8 and 25. Those are searches where Google already associates you with the topic but you are not visible enough to get the click. A dedicated page, or a section added to an existing one, moves these faster than anything you build from scratch. This is the cheapest work available and it is sitting in every account I have ever audited.

Then look at the page-level view for your highest-value URLs and check which queries they attract that you did not intend. A pricing page ranking for “how much does X cost for enterprise” is telling you to build an enterprise pricing page.

One caution: Search Console anonymises low-volume queries, so a meaningful share of your long tail is invisible in the interface. The API returns more than the UI does, and comparing total clicks against the sum of listed queries will tell you how much is hidden. On small B2B sites it is frequently 40% or more.

Read your competitors’ page inventory, not their keyword list

Keyword tools will show you what competitors rank for, filtered through the same broken volume data. More useful is what they have chosen to build.

Crawl their sitemap. If a competitor has 240 integration pages, someone there ran the numbers and decided that pattern was worth engineering time. If they have a page for every US state, they are chasing local modifiers. If they built out use-case pages for eleven verticals and then stopped, look at which eleven and ask why those.

This tells you about demand patterns rather than individual terms, which is more useful at the roadmap stage anyway.

Sorting what you find

Once you have a raw list from these sources, the sorting matters more than the collecting. We score every cluster on four things: how close it sits to a purchase decision, whether we can realistically rank for it within a year, roughly what a visitor from it is worth, and whether we have anything genuine to say.

That last criterion kills more clusters than the other three combined. A keyword you can rank for, with buying intent, in a topic where you have no distinct point of view, will produce a page indistinguishable from four other pages. Skip it.

Here is the shape of the intent tiers we use, roughly in descending order of value per visitor:

Someone searching your competitor’s name plus “alternative” is actively trying to leave. Someone searching “[category] for [their industry]” has decided the category is right and is filtering vendors. Someone searching “[category] pricing” is building a business case. Someone searching “[your product] vs [competitor]” is on a shortlist. Someone searching “how to [do the thing your product does] in Excel” is a year out but will remember you. Someone searching “what is [category]” is probably writing a college assignment.

Most content programs invert this list.

A note on volume you should chase

I am not arguing you should never build top-of-funnel content. I am arguing it should be a deliberate minority of the plan with a stated job, usually earning links or building topical depth around a commercial cluster that needs it.

The test we apply: can you name the money page this article links to, and would you be happy if the article ranked first and nobody ever converted directly from it? If yes, build it. If the honest answer is that you hope some of the traffic converts, you are building on hope.

What a finished map looks like

For a mid-market SaaS company, the output of two or three weeks of this work is usually 60 to 140 keyword clusters, each with a target URL (existing or planned), an intent tier, an estimated annual value, and a note on what makes our angle different. Total search volume across the whole map is often under 20,000 a month, which looks unimpressive next to what a generalist agency will show you.

The generalist’s map will have 800 keywords and 400,000 monthly searches and about eleven terms anyone would ever buy from.