Write the question, name what would answer it, then search. Deciding what counts as evidence afterwards is how bias enters unnoticed.
Do now: Scope one real question and write its stopping rule before opening anything.
A stopping rule is what turns research into a finished piece of work.Research without a stopping rule expands until the deadline arrives, and research without an evidence standard produces whatever the first few sources happened to say. Both are common, and both are fixed at the start rather than the end.
Sharpen the question first
A researchable question has a subject, a scope and a time frame. 'Is solar worth it' is a mood. 'For a household in Nyeri spending about this much on electricity monthly, what is the payback period on a system of this size, at current prices' can actually be answered, and it tells you exactly what to look for.
Then write what would count as an answer. A number with a range. A named regulation with a date. Three practitioners saying the same thing independently. Deciding this in advance is the single most effective guard against finding what you hoped to find.
AI is genuinely good at this stage. Ask it to restate your question more precisely, name what evidence would settle it, and list what it cannot establish. That is pattern work, and it is where the tool is strong.
Decide what would change your mind before you look. Afterwards, everything is evidence for what you already think.
Search wide, verify narrow
Use AI to widen the search: what terms are used in this field, which institutions publish on it, what are the main positions, what would a sceptic ask. Widening is low-risk because you will check everything anyway.
Then narrow hard. Pick the three sources that would most change your answer and open those properly. Prefer origins over retellings — the regulator's publication, the original study, the actual price list. Every layer of retelling loses conditions.
Record four things per source: what it says, where it is, when you accessed it, and how confident you are. The confidence column is the one that makes the research honest, because it forces you to admit which parts are still soft.
Know when to stop
Write the stopping rule with the question: 'I stop when I have two independent sources agreeing on the payback range, and I have checked one against actual local prices.' Without one, research ends when you run out of time, which means it ends wherever you happened to be.
Stop early too when the answer is that the evidence does not exist. That is a finding, and reporting it honestly is far more useful than assembling weak material into the appearance of an answer.
Finish by writing what would change your conclusion. It keeps the work honest, it tells a reader how much weight to place on it, and it makes revisiting the question cheap when something new appears.
'The evidence does not exist' is a legitimate result. Manufacturing an answer is not.