Feedback from readers is one of the most useful and most misused inputs available. The useful part is that people reliably notice when something is not working. The misused part is taking their explanations at face value, since readers are describing an experience rather than diagnosing a cause, and the two are frequently unrelated.

The old formulation is that readers are almost always right that something is wrong and almost always wrong about what would fix it. Working from that assumption makes feedback considerably more valuable and considerably less painful.

Location is the signal

When several readers say a chapter dragged, the chapter probably dragged. When they say it dragged because there was too much dialogue, that is a guess, and the actual cause might be that the scene had no decision in it, which is the structural test in the piece on building a scene from one decision.

So record where comments cluster and treat the explanations as raw material rather than as findings. A cluster of unrelated complaints about the same section is strong evidence that the section has a problem, whatever anybody thinks it is.

Independent repetition matters

One person's reaction is information about that person. Several people arriving at the same reaction independently, without having seen each other's comments, is information about the work. Distinguishing the two is the single most useful filter available.

Comment threads make this harder, since the first comment shapes the rest. Feedback gathered privately, or from readers in different places, is more reliable than a discussion where everybody has read the same opinion first.

Requests are the weakest input

Specific requests for what should happen next are the least useful category, because readers are describing what they would enjoy in the moment rather than what would make a satisfying work. Giving an audience exactly what it asks for reliably produces something nobody remembers.

That does not mean ignoring them. A pattern of requests indicates what an audience is currently interested in, which is genuine information about where the work has created expectations, and expectations are worth knowing about even when they should not be met directly.

Praise is information too

Which parts get quoted, screenshotted and shared tells you what is working, and creators consistently pay less attention to this than to complaints. Frequently the answer is surprising: a quiet page rather than a dramatic one, a minor character, an aside.

Following that signal is how a work finds what it is actually good at, which is often not what its maker planned. It is also more actionable than negative feedback, since it identifies something to do more of.

Who the feedback is from

Feedback comes disproportionately from the most engaged readers, who are not representative of the whole audience. They read closely, care about continuity, and want more of everything, and optimising for them produces work that is dense and inaccessible to newer readers.

This is the audience-composition question raised in the piece on who actually reads this. Knowing that the loudest readers are a specific slice of the audience is part of reading their comments correctly.

Protecting yourself while reading it

Reading feedback immediately after publishing, when you are tired and exposed, is the worst possible time. Collecting it and reading it in a batch, later, at a set time, produces better decisions and considerably less damage, which connects to the sustainability argument in the piece on burnout.

It is also entirely legitimate to have somebody else filter it. Many working creators do, and it removes the small proportion of genuinely abusive material without losing anything useful.

Silence deserves reading too, and it is the hardest signal to interpret. A chapter that produces no reaction at all is frequently a worse sign than one that produces complaints, because complaints require engagement and silence does not. Tracking which instalments generate nothing is a crude but useful measure of where an audience quietly disengaged.

Reading feedback so it improves the work

Trust the location of complaints and distrust the diagnosis, weight independent repetition heavily, treat requests as information about expectations rather than instructions, follow what gets shared, remember who is speaking, and read it in batches at a chosen time.

Letters to this journal are welcome through the letters page.