How to Measure Share of Search Across a Competitive Set

Share of search is the proportion of all branded searching in a category that goes to one brand. It is built from what buyers typed, which makes it the rare competitive measure that reports on demand rather than on effort, and the free version of it costs nothing beyond the time to assemble it.

The contrast worth drawing immediately is with the metric it is most often confused with. Share of voice counts what companies emit: advertising money, mentions, coverage, presence. Share of search counts what happened next. During a large campaign the two can move in opposite directions, and that gap is itself the finding, because it says the spending did not translate into anybody looking you up.

Why it belongs in competitive measurement at all

Most market-position numbers are either unavailable or stale. Nobody counts business software categories, published shares arrive a year late, and survey-based awareness tracking is expensive enough that most companies run it annually if at all. Search volume for a set of brands is available this afternoon, it is the same instrument for you and for every competitor, and it updates weekly. That combination is the entire practical case for the metric, and it stands whether or not the predictive research further down does.

One country, one window, one tool

All three have to be held fixed, and the third catches people out. Search volumes from different tools are not comparable, because one reports a relative index built from a search engine’s own query data and another reports a modelled monthly figure derived from clickstream panels. Mixing them inside one calculation produces a share that is partly an artefact of which brand you happened to look up where. Finding a competitor’s keywords works through what each tool is actually measuring, which is worth reading once before you commit a series to one of them.

Comparing more than five brands at once

The free trend tools cap a comparison at five terms, and most competitive sets are larger than that. The way round it is to chain overlapping batches through a shared anchor brand. Compare brands one to five. Then compare that same anchor brand against brands six to nine over the identical date range. Multiply every value in the second batch by the ratio of the anchor’s index in the first batch to its index in the second, and all nine brands now sit on one scale.

Two lists determine this metric before any division happens: which brands are in the set, and which queries count as a search for a brand. Both are decisions, both are usually made once and never revisited, and between them they matter more than anything else on this page.

The brand list is the easier of the two to get right and the easier to get wrong quietly. Include every name a buyer would realistically put on the same shortlist, including the ones that irritate you and the ones you consider too small to matter. Omitting a genuine competitor raises everybody’s share, yours included, and turns the series into a record of your own list-keeping.

The brand name alone under-counts, and measurably so

A basket built from bare company names is the default choice and it is too small. Kantar measured the gap in August 2025 and found that widening past the bare name lifted recorded interest by 36% on average. The uplift lands exactly where you would expect it, on companies whose product is more famous than they are.

That figure is not a benchmark and cannot be applied to your own numbers. What it does establish is that the gap between a lazy basket and a careful one is large enough to change conclusions, and that it varies by brand, which means it does not cancel out when you divide.

Brands that are ordinary words

Several well-known software companies are named after common English nouns, and their raw search volume includes everybody who meant the word. Trend tools handle this through an entity rather than a string: select the suggested company or subject rather than the plain search term, and the tool resolves the meaning for you. Where no entity exists, the workable substitute is a compound query that cannot mean anything else, such as the brand paired with the category noun, applied identically to every brand in the set so the comparison stays fair.

The analysis that put this metric on the map is explicit about something most write-ups of it leave out. The link between searching and buying runs in both directions: people search for brands they might buy, and people search for brands they already own. Both kinds of query land in the same volume figure, and nothing in the arithmetic separates them.

The consequence is systematic rather than random. A competitor with a large installed base collects a steady daily stream of people typing their name to reach a login screen, a status page or a help article. A challenger with few customers collects almost none of that. So branded volume flatters whoever is already biggest, by an amount proportional to exactly the thing you were trying to measure independently.

Two layers, read separately

The fix is to stop treating branded search as one number. Split the basket into a navigation layer, which is the brand paired with login, support, status and the rest, and an evaluation layer, which is the brand paired with versus, alternatives, pricing and reviews. Compute a share for each.

The two shares answer different questions and the distance between them is the finding. A company whose evaluation share is well above its navigation share is being considered more often than its customer base would suggest, which is what a challenger gaining ground looks like before anything reaches a revenue figure. The reverse pattern, a large navigation share and a thin evaluation share, describes a company living on its existing customers.

What this does to the headline number

Excluding navigational queries will usually lower the leader’s share and raise everybody else’s, which makes the corrected series look worse for an incumbent and better for a challenger than the uncorrected one. That is not a thumb on the scale, it is the removal of one. Decide the rule before you look at the result, write it into the sheet, and apply it to every brand in the set including your own.

This is one of the few competitive metrics with real published work behind it, which makes it worth reading the findings precisely rather than through the version that circulates.

What the 83% figure is and is not

It is an average relationship observed across 30 case studies, published by a trade body which noted in the same release that nothing in the data shows one of the two figures causing the other. It is not a conversion factor, so multiplying your share of search by any number to obtain a market share is an operation the evidence does not support. Nor is it a target, since a figure describing an average across twelve categories says nothing about where your brand should sit.

The lead times follow the purchase cycle

The original analysis found search movement preceding share movement by up to a year in car buying, around six months in mobile handsets and about three in energy. Those are not three results, they are one result: the lead time is roughly how long a purchase decision takes in that category. Read that way it is a genuinely useful finding, because it tells you what lag to expect in your own market instead of giving you somebody else’s number to borrow.

Why a business software category is a harder case

All three studied categories share a property that considered business purchases do not: one person searches and that same person buys. A software evaluation involves several people searching on behalf of one decision, at different depths and different moments, and a meaningful share of the evaluation happens in places a search engine never observes, such as peer conversations, analyst calls and procurement shortlists drawn up in advance.

None of that makes the metric useless here, and it does change what you should claim for it. Treat the published work as evidence that the mechanism exists, then measure the lag on your own category by lining up your share of search against your own bookings for as many quarters as you have. If no lag shows up, you have learned something worth knowing about how your buyers behave.

The most useful finding in the original work is also the least quoted. Plotting share of search against share of market brand by brand produced a similar slope for every brand and a different height for each one. The relationship holds, and each brand has its own base level inside it.

That single observation removes most of the ways this metric gets misused. It means there is no level that is good in the abstract, no cross-brand comparison of the two figures that means anything, and no conclusion to be drawn from your share of search sitting below your market share. Plenty of perfectly healthy brands sit there permanently, for reasons that include how much of their business renews quietly and how much of their category searches by product name instead.

Use a rolling average, not the month

Branded search is seasonal, it reacts to news, and a single funding announcement can double a small brand’s volume for a fortnight. A twelve-month rolling average absorbs all of that and still turns over monthly, which is why it behaves like a trend while the raw month behaves like noise. Report the rolling figure and keep the monthly one underneath it for investigating a spike.

The sheet, and the columns that make it auditable

One row per month, one column per brand, then the set total, your share, and a short note on anything unusual. Above the table, four fixed lines that never change without a deliberate decision: the country, the tool, the query rule for what counts as branded, and the list of brands with the date each was added. A share of search series without those four lines cannot be interpreted by anyone who did not build it, including you in a year.

The data comes out as an export. Trend tools offer a CSV download of the exact comparison on screen, keyword tools export volumes per term, and either way the monthly job is a download and a paste rather than a research exercise. Keep the raw export beside the sheet: when a month looks wrong, the question is almost always whether the query set changed, and only the raw file can answer it.

Giving each month somewhere to be explained

The note column is what turns this from a chart into an analysis, and it can only be filled in if somebody was watching at the time. A competitive intelligence platform already tracking the companies in the set makes that column cheap: what each one announced, launched, renamed or spent money on arrives with a date attached, so the note comes from a record rather than from somebody’s memory at quarter end.

Two other series belong alongside this one and are worth pulling at the same time, since the same tools produce them. Competitor website traffic shows whether the searching turned into visits, and competitor ad spend shows whether a rise in a competitor’s branded volume was bought rather than earned.

Every other failure on this page is a decision made badly. This one is a decision made correctly that stops being correct, and it is specific to a metric built on strings of text. A competitor rebrands. A product is renamed and the old name stops being typed. Two companies merge and one name is retired into the other. A sub-brand launches beside the parent and takes half the volume with it.

In each case the company is still there, still selling, still winning deals, and your basket is counting a word people have stopped using. The series records a collapse in their share and a rise in yours, and the sheet gives no sign that the market itself held perfectly still. The same mechanism runs the other way when a competitor launches a brand campaign: the denominator swells for a quarter, your share falls, and no buyer became less interested in you.

Catching either one requires knowing what those companies did in the month the line moved. Flares follows launches, renames, acquisitions, campaigns and messaging changes across the companies you track and keeps them dated, so a basket can be corrected when the name changes rather than three quarters later when somebody questions the chart.

What monitoring will not do is supply the volumes. Those come from a search tool, they are modelled or indexed rather than counted, and no amount of change detection improves them. The half that can be fixed is knowing which words to count and what happened in the month they moved, which happens to be the half that breaks.