Two minute read

Writing a topic that works

PaperPatrol works exactly as well as the sentence you give it. This page is the whole of what makes one sentence work and another one fail.

The test

Before any advice, try this. It takes about fifteen seconds and it is the whole of the method.

Picture this. A new colleague starts on Monday. You hand them a stack of papers from the last two weeks and one job: pull out the ones that matter to you. Then you leave for a conference. They get one sentence from you, and no chance to ask a follow-up question.

What sentence do you give them?

Whatever you just thought of is what you give PaperPatrol. Not your keywords, not the title of your project, not the name of your field. The sentence a person could actually sort a pile with.

The test works because it moves the problem somewhere you already have good instincts. You would never hand that colleague the word "cancer" and expect a useful pile back. You would tell them what you are chasing, in what system, and why a paper counts. Everything below is that instinct, written out.

And the reverse holds. If your sentence would leave that colleague guessing, it leaves PaperPatrol guessing too, and a guess arrives in your inbox as a long list you have to read yourself.

What happens to your sentence

It is used twice, and the two uses pull in opposite directions. Knowing that is most of the skill.

First it is used to search. PaperPatrol works out which terms to ask PubMed and bioRxiv for, including near neighbours you did not write, so a paper that calls your subject by another name still surfaces. Your sentence sets the neighbourhood.

Then it is used to judge. Every abstract that came back is read and held against your sentence, one at a time, and only the ones that genuinely match are flagged.

This is why a vague sentence fails in a way that feels backwards. It does not return too little. It returns too much, because a sentence that describes a whole field gives the judging step nothing to reject with. Almost everything matches "neuroscience". You get a long list, and sorting a long list is the exact job you signed up to stop doing.

The four parts

A sentence that works almost always names three things, and sometimes a fourth. You do not need the labels, only the habit of checking that each one is there.

The subject What thing CAR-T cells · callosal projection neurons · gut microbiome

The process Doing what, or what happening to it exhaustion · axon guidance · off-target editing

The context Where, in what system, in what disease pancreatic cancer · developing cortex · Parkinson's disease

Optional The method or model, only if it decides relevance in mouse · single-cell RNA-seq · live imaging

Put together, with the parts still visible:

CAR-T cell exhaustion mechanisms in pancreatic cancer immunotherapy

Add the fourth part only when a paper using a different method would be useless to you. If you would still want to read it, leave the method out: it narrows the search without making the judgement any sharper.

The same topic, three times

cancer

One part out of three. A subject and nothing else. Thousands of papers a week are about cancer and almost none of them are about your question. Every one of them matches, so you receive a digest you have to triage yourself.

T cell exhaustion in cancer

Two parts out of three. Subject and process are there, the context is still an entire discipline. Better, and still a weekly list that spans every tumour type and every therapy.

CAR-T cell exhaustion mechanisms in pancreatic cancer immunotherapy

All three. A paper about T cell exhaustion in melanoma now has a reason to be rejected, and a paper about CAR-T persistence in pancreatic tumours has a reason to be flagged even though it never uses the word "exhaustion".

Too narrow is a failure too

It is the rarer mistake, and it is quieter, which makes it worse: nothing arrives, and nothing arriving looks exactly like a calm week in your field.

A sentence is too narrow when it names one gene, in one species, with one technique, and would reject the same finding shown any other way. If you would want to read a paper that reached your result from a different direction, your sentence needs to leave room for it.

A rough calibration: over a couple of months, a well aimed topic tends to bring something most weeks and nothing on some of them. Nothing for six weeks running usually means the sentence, not the field.

Habits to keep, and to drop

Write it like this

A plain phrase, the way you would say it out loud.
Spell the subject out the first time, then the abbreviation is welcome too: long QT syndrome (LQTS).
Name the disease, tissue or system you actually care about.
Keep it to roughly one line. Two clauses is plenty.

Not like this

No boolean syntax. AND, OR, NOT, quotes and brackets are read as words, not as operators.
No keyword lists. cancer, immunology, T cells is three broad topics, not one narrow one.
No author or journal names. Relevance is judged on the science.
No questions. What causes ALS? describes a field, not a filter.
No bare abbreviation on its own: CAR is a receptor, a vehicle and a gene, depending on who is reading.

Check your sentence

Read your sentence aloud. Can you hear the subject, the process and the context, each one separately?
Name a paper in your field that is close to your topic but that you would not want to receive. Does your sentence give a reason to reject it? If not, add the part that would.
Name a paper you would hate to miss. Would your sentence catch it if the authors used different vocabulary for the same thing?
Count the fields you have named. One is too few. Three is right.

After your first digest

The first scan covers the last 14 days, which is the fastest feedback you will get. Read what it flagged and judge the sentence, not the papers.

Too much arrived, and it is loosely related? Add the missing context, the disease or the system. Nothing arrived at all, two weeks running? Take out the method, or widen the context by one step. You can rewrite your topic whenever you like from your personal area, and rewriting it deletes nothing that has already been found for you.

That is the whole of it. The sentence is the product.

Write mine now