The question is not how many, it is what arrived

A name, an email address and a message the visitor typed, plus the time it arrived and whether the notification went out. The form takes those three fields and no more, and the message is the visitor's own words rather than anything the assistant composed.

There is also a hidden field on the form that no person ever fills in. A submission carrying it is answered as though it succeeded and is not stored, so the obvious automated junk never reaches the list. That keeps the list clean and it also means nothing counts what was dropped.

The conversation the enquiry came out of is recorded against it when the identifier the browser sent genuinely belongs to that assistant, which is checked rather than trusted. So the link exists in the record. The enquiries screen does not display the conversation, and the notification email does not carry it.

Consider when it is written. Somebody has just failed to get an answer, they have been offered a form, and they are now retyping their problem for the second time into a smaller box. Almost nobody restates the context they already gave. Most people write a compressed version of the question they already asked, or something even shorter, because from where they sit they have said it once already.

So the systematic quality problem is not carelessness on the visitor's part. It is that the thing that would make the enquiry answerable is sitting in the conversation immediately before it, and the message is the one place it does not need to be repeated for the visitor and is the only place it exists for the reader.

Knowing that shapes what you do about it. The lever is not asking visitors to write more. It is deciding, on your side, to read the conversation before replying, and building that into how enquiries are handled.

Nothing in the product judges an enquiry. There is no score, no flag for one that looks incomplete, no grouping by topic, and no record of what happened after it left. The reply is written in your email, which is a place this product cannot see, so it cannot know whether the answer took one message or four.

The measurement that works is a tally kept by the person who answers them. Two columns: answered from what arrived, and had to ask what they meant. A month of that is a real number about a real cost, it takes seconds per enquiry, and it points at something you can change.

There is one field that is reported, and it is worth checking first for a different reason. Each enquiry records whether the notification email was delivered. An enquiry nobody received has no quality at all, and it is the failure most likely to be invisible.

Add a required field asking what it is about. Every enquiry that arrives now has a category, the list looks tidier, and fewer of them arrive at all, because each extra field loses people. The ones you lost were disproportionately the ones in a hurry, which is to say the ones with something urgent.

Judge quality by how quickly enquiries were answered. Your team will answer the easy ones first and the figure will improve while the difficult ones sit. This is the oldest distortion in support and it does not stop being true because the queue is an inbox.

The accidental version is a change to the fallback wording. Reword the refusal so it points more firmly at the form and enquiry volume rises, mostly with people who would have found the answer given one more try. The list is longer, each item is thinner, and nothing about the material changed.