AnswersBusiness software

How do you find out whether AI assistants recommend your software?

An operations lead types the job into an assistant and reads back a list of tool names. That list is the shortlist now, and it is built before anybody visits a vendor site. Measuring where you sit on it means asking the questions prospects ask, across the assistants they use, and reading what comes back about your plans, your integrations and the route to a trial.

Where does a software shortlist get built now?

Increasingly inside an assistant, before anybody visits a vendor site. A prospect describes the team, the stack and the constraint, and reads back a set of names that becomes the list they evaluate.

An assistant asked that question composes an answer out of what has been written about the category and whatever it retrieves at that moment, and what comes back is a set of tool names in an order. For business software that set is the shortlist: the list a company works through before anybody books a demo or starts a trial.

Which means the evaluation starts a step earlier than most software teams measure. By the time a prospect reaches a pricing page they have already narrowed the field, and the narrowing happened somewhere nobody was watching.

So the useful question is not how much traffic an assistant sends. It is whether you are named at all when the job you do is described, and whether what is said about you is right.

Can an assistant route a prospect to a trial, and does that path work?

That is the last step for software, and Delphi calls it Signed up. It reads whether an assistant sends somebody to a page where they can start a trial or book a demo, whether it describes your plans and your integrations correctly, and whether those routes resolve.

The step reads the hop that replaces a checkout. The assistant names you, and then it has to hand a prospect somewhere they can act: a trial they can start, a demo they can book, documentation they can read.

Three things decide it. Whether the route an assistant offers resolves. Whether your plans are described as you publish them. And whether your integrations are, because integrations are one of the constraints a prospect types alongside the job itself.

A recommendation that ends at a home page with no next step is a recommendation that stops.

Does it matter where the prospect is sitting?

An assistant places software by the job it does and the stack it fits rather than by a map, so the reading follows the tools a prospect already runs and the shape of their team rather than a catchment or a set of opening hours.

A clinic is found by where it is, and a venue is found by where it is. Software is found by what it does, so the reading is built out of the job rather than out of an address, a catchment or a set of opening hours.

What takes the place of the map is the stack and the shape of the team. A prospect describes the tools they already run, the size of the team, the certification they need and the way they want it deployed, and the assistant places you against that.

It also changes who is reading. Business software is evaluated by a company rather than picked by one shopper, so a demo booked or a place on a shortlist counts alongside a trial started, and all three are read as the same last step.

Which sources decide whether you are on the list at all?

Software review platforms and comparison sites carry the most weight, alongside practitioner threads and developer forums, with your own documentation and pricing pages cited directly in answers about how you work.

Review platforms and comparison sites are the first of those, because they are written to answer exactly the question a prospect is asking. Practitioner threads and developer forums sit beside them rather than below them: they are where a tool gets recommended by name, by somebody who runs it.

Your own pages are the only source you fully control, and documentation and pricing pages are the ones quoted directly in answers about how you work. Trade and technology press sits behind those, and it is what gets cited when a category roundup is being written.

So the mix is read across those sources rather than off a shelf, and the two that carry the most weight, the review and comparison platforms and the practitioner threads, are both places where somebody else writes about you.

What do prospects actually ask before they shortlist?

Three families carry most of it: alternatives and head to heads, the job to be done phrased as the best tool for a named team or workflow, and the broad category question that decides who is on the list at all.

Alternatives questions lead. Somebody names a tool they already run and asks what else does this, which is how most software shortlists are built.

Next is the job to be done, usually phrased as the best tool for a named team or a named workflow. Then the broad category question, which settles who is on the list at all before any comparison starts.

Underneath those sit the constraints a prospect types alongside: integrations, deployment, certification, team size. And the how-to and definition questions, asked before the decision rather than at it, which are where the language of a category gets set and where a vendor can be the one who sets it.

Does a public API or an MCP server change what an assistant can do with you?

Yes, and it is the one reading written specifically for software. A public API or an MCP server is a real route rather than a forward-looking one, so it is read as part of the last step rather than as something to think about later.

For a practice, a venue or a component maker the machine route belongs to somebody else: a booking system, a platform booking API, a parts catalogue. For software it is your own product, because the product is already an interface, and an assistant that can call it can do more than describe it.

So the last step reads it as a route. Beside it sits the plainer reading, which is whether the trial, demo and documentation links an assistant offers resolve when somebody follows them.

The path is also what stands in place of the retail checks. A catalogue line and a shelf belong to a shop, so what gets read here is whether the route an assistant offers works and whether there is a machine route beside it. That is the version of readiness that means something for software.