How a Market-Size Figure for This Category Gets Made
A market-size figure for AI companion apps is not an observation. It is an estimate manufactured by a specific method, sold as a product, and promoted with a free headline number. No figure appears on this page, because supplying one would mean either inventing it or laundering somebody else’s assumptions — and the second is how most of the numbers in circulation came to exist.
Knowing how they are made is more durable than knowing any of them.
Who publishes them, and why
Market-research vendors produce reports and sell them. The headline number is the marketing for the report, released free precisely so it travels. The method is inside the paid document, or nowhere.
Companies raising money need a large addressable market, because that is what the pitch requires. Figures in a funding announcement are chosen to support a case.
Trade and technology press need a number to anchor a story, and the free press release supplies one at no cost.
None of that requires bad faith. It does mean every figure you will meet was published by someone who wanted it to be large.
How the estimate is actually built
The common approach is top-down, and it is worth seeing spelled out because it is less impressive than the output looks.
Start with a bigger market whose size is thought to be known — consumer apps, or subscription software, or a broader category of conversational products.
Assume a share of it belongs to this category. This is an analyst’s judgement.
Assume a number of users and an average amount each spends. Both are estimates, and the user-count problem alone is severe enough to sink the exercise — see why nobody can tell you how many people use companion apps.
Multiply. The result inherits every assumption above, compounded, with no error range attached by the time it reaches a headline.
Then project it. A growth rate is selected and applied forward to a chosen year. That final figure is the one that gets quoted, and it is the least evidenced number in the whole chain: it is an assumption applied to an estimate built on assumptions.
Why a projection is an artefact rather than a finding
The growth rate is chosen, not measured. Small changes in it produce enormous differences at the far end.
The horizon is chosen for comfort. Far enough away that nobody will check, and near enough to sound relevant.
Nobody revisits. Projections from previous years are not scored against what happened, so there is no feedback and no accountability. This is the structural reason the genre does not improve.
Different vendors publish very different numbers for the same year, which is the clearest available evidence that none of them is measuring anything. When you meet one figure, look for a second; the spread is the real information.
The citation loop
This is how a manufactured number acquires the appearance of a fact.
A press release states the figure. An article reports it, attributing it to the vendor. A second article cites the first. A third cites the second, by which point the vendor, the method, and the date have all dropped out. Eventually the number appears with no attribution at all, as background — and it is now unfalsifiable, because there is nothing left to check.
The same laundering happens to usage and behaviour statistics, and the general defence against it is set out in how to read a statistic about this category.
The tells, in rough order of usefulness
No stated method. If the article does not say how the figure was produced, it does not know.
Improbable precision. A category with no agreed definition cannot be sized to several significant figures, and precision in that context is a presentation choice.
A distant horizon with a confident tone. “Expected to reach”, “projected to grow to”, by a year far enough out to be safe.
The figure appears only in secondary coverage. Follow it back. Frequently the trail ends at a press release.
No date for the underlying data, as distinct from the publication date.
No definition of the category. Which, given how the boundary problem works, means the figure could refer to almost anything.
What a market figure could not tell you anyway
Even a perfect one would be irrelevant to every decision a reader of this site is making. The size of a market says nothing about whether a particular app handles your conversations acceptably, what it charges, what it does when something goes wrong, or whether your own use of it is doing you any good. It is information for investors, and it is only sometimes information for them.
Where the same numbers are put to work justifying predictions about the category, the genre has its own problems — described in why a trends piece is a genre rather than a finding.