A Search Box Should Not Pretend to Be an Answer
By The Paragraph
The convenience of a fluent reply comes with an obligation to show where its confidence ends.
Type an awkward question into a search box and a neat paragraph can feel like a small act of rescue. The competing pages, caveats and unfamiliar terms have disappeared. Here is an answer, already dressed for the occasion. The difficulty is that good tailoring tells us very little about what is underneath.
A list of sources makes its untidiness visible. We can see that different organisations have different interests, that one account is older than another, and that the answer may depend on which question we mean. A generated response can compress those differences into a single voice. That voice may be helpful. It may also make uncertainty unusually difficult to hear.
The objection is not that people should endure inconvenience for its own sake. A clear explanation can open a subject to someone who would otherwise abandon it. Translation, summarisation and help with unfamiliar vocabulary can be valuable. The design question is whether the service gives readers a path from its convenient explanation to the evidence that ought to support it.
Imagine asking why a local planning application was refused. An answer might blend an official decision, an applicant's statement and an angry discussion among neighbours. Each may contribute useful information. They do not have the same status. The reader needs to know what the authority decided, what somebody alleged, and what the system has inferred between the two.
A row of links is not enough if those links merely decorate claims they do not establish. The important connection is between a particular statement and the source that supports it. Dates matter where circumstances change. Disagreement matters where the evidence remains contested. An answer should make those features easier to inspect, even if that costs it some rhetorical smoothness.
There is a reasonable concern about overwhelming readers. Every ordinary explanation leaves things out; no interface can reproduce an entire library on a phone screen. But layers are possible. A short answer can state its limits and let the curious reader open the relevant passage. Brevity need not mean that the distinction between knowledge and guesswork has been tidied away.
We should also be wary of making users responsible for detecting every plausible error. Telling people to check everything defeats much of the promised convenience. If a product presents itself as saving the work of investigation, it assumes some responsibility for indicating when investigation remains necessary. Confidence should follow evidence rather than the grammatical completeness of a sentence.
Readers have a part to play nonetheless. For consequential questions, it helps to ask what kind of answer is being offered: a calculation, a summary, a recommendation or a prediction. These deserve different forms of trust. The cheerful certainty of the interface should not persuade us that they are all variations of the same simple transaction.
Consider, too, the difference between missing evidence and evidence of absence. A system unable to find a relevant record should not convert that failure into a confident claim that no record exists. The honest description of a search sometimes includes what was searched and what remains beyond its reach.
The best search assistance would leave people better able to judge what they have found. It would sometimes answer directly, sometimes expose a disagreement, and occasionally admit that the question lacks enough information. A machine that can produce a paragraph is impressive. A service that helps us know when to believe one is considerably more useful.