Most of a research brief is not findings. It is constraints. The sections that change a recording most are the list of statistics the host is told not to use, the section arguing against the episode's own premise, and the appendix of claims that could not be verified against a primary source. All three arrive before the host sits down, which is the only point at which they cost nothing.
People tend to picture a research brief as a summary. A tidy document with the main points on it, something to glance at while you talk.
Mine reads more like a set of instructions about what not to say.
A recent brief I built, on why people feel lonely in a hyper-connected world, runs to about 390 lines. The findings, the part everyone assumes is the whole document, take up maybe a third of it. The rest is scaffolding for one job: making sure the host does not say something on air that cannot be defended six months later.
Three parts of it change what gets said on the day more than anything else in there: the numbers the host is told not to use, the strongest case against the episode, and an appendix of claims that could not be traced back to a primary source. The search side of the brief is a different subject, and I have written about it separately.
The part that does the most work is the list of numbers you are not allowed to use
Every brief carries that list, and it is exactly what it sounds like. Figures that are circulating widely, that would land beautifully in an episode, and that will not survive being checked.
The loneliness brief quarantined thirteen of them. An insurer-produced index built on a self-selected online panel, useful for colour and useless as prevalence. Three separate agency-produced "percentage of Gen Z who are lonely" figures with opaque methodology. Market forecasts for AI companion apps, published by vendors about their own market. A claim that loneliness shortens lifespan by fifteen years, weakly sourced and circular.
And one worth walking through properly, because it shows the mechanism.
The comparison between loneliness and smoking is everywhere. It traces to a 2010 meta-analysis by Julianne Holt-Lunstad, Timothy Smith and J. Bradley Layton in PLoS Medicine, covering 148 studies and 308,849 participants. What the paper reports is a 50% increased likelihood of survival for people with stronger social relationships, and it says the influence of social relationships on mortality risk is comparable with well-established risk factors. That is a real, large, carefully qualified finding.
The cigarette number is a translation of it that happened downstream. By the time it reaches a podcast it has usually become a flat clinical equivalence, stated as fact, with no mention of the effect-size comparison it started as and no mention of the later cohort work that complicates it.
The brief does not ask the host to drop the claim. It gives them the wording that keeps it true. Frame it as an effect-size comparison, and voice the counter-finding out loud in the same breath.
That instruction is close to useless once a recording exists. Before one, it costs a sentence.
Part of every brief is written against the episode on purpose
That part's only job is to make the strongest available case that the episode's premise is wrong.
For the loneliness episode, it argues the adult epidemic is overstated and inconsistently measured, that prevalence estimates swing by more than ten points depending on question wording, that economic precarity and declining civic institutions may be driving what gets attributed to loneliness, and that the signal which holds up is concentrated in teenagers and specific groups rather than the whole population.
It also states the incentives in both directions. Media benefit from crisis framing and the wellness industry benefits from selling solutions, and equally, platforms and sceptics have their own reasons to minimise. Naming only the first half is itself a bias, and the audit catches it when I do that.
It ends on a distinction I now put in every brief: which parts of this are values and which are contestable facts. "Human connection matters for health" is well supported and also a shared value. "Loneliness is an epidemic rising rapidly for all adults" is a factual claim that is genuinely in dispute. The episode asserts the first confidently and hedges the second.
A host who has read that has already met the best version of the objection, in writing, from someone paid to find it.
The appendix is the part nobody expects
Before the brief reaches anyone, it goes through a separate check whose whole job is to find what I got wrong. Whatever cannot be traced back to a primary source goes into an appendix at the back, with a status and the exact document that would resolve it.
Thirteen entries in the loneliness brief. A three-decimal effect size that could not be located in any cited source and reads as invented precision, cut entirely and replaced with the qualitative statement the underlying trials actually support. A lawsuit that is well documented and a platform policy change that is well documented, with the causal link between them removed, because only the two facts are established. A prevalence figure kept with its decimal flagged, because the number is plausible and the primary PDF had not been fetched. A named person in a news segment, tagged do not name on air until confirmed.
None of that reaches the microphone as a confident sentence. It reaches it as a flag, or it does not reach it at all.
What this changes on the day
Three things, and I want to be precise about which of them I can evidence and which I cannot.
The first is that the host stops improvising numbers. Every figure they might reach for has already been graded, and the ungradeable ones are visibly quarantined rather than quietly absent, so there is no gap left for memory to fill.
The second is that the objections arrive early. The strongest counterargument is in the document, so the episode deals with it as a matter of course instead of the host meeting it in a reply three weeks later.
The third is the one I am least able to prove. A host who knows in advance how the argument runs and how strong each piece of evidence is speaks with a different kind of confidence than one working from a topic list, and I believe that shows up in the recording. I have not measured it against a control, and I am not going to pretend a comparison exists that does not.
I have no controlled comparison between episodes produced with this kind of brief and episodes produced without one. As far as I can find, nobody has run that study in a podcast context. What I can show is the document, the audit that checked it, and the specific claims that were removed and why. That evidence is auditable. A performance claim on top of it would not be.
I also do not assess the medicine. Where a health claim appears in a brief, my finding is about its sourcing and whether the confidence of the sentence matches the strength of the study behind it. Whether the underlying clinical picture is right is the practitioner's call, not mine.
The findings are the easy part. Any competent researcher can assemble a topic summary, and many hosts can do it themselves.
The work that is hard to do for yourself is the adversarial part. Deciding which of your favourite statistics has to go, writing the best case against the thing you are about to record, and marking the claims you would rather were true as unverified. It is uncomfortable, and it is much cheaper before a recording than after one.