What Strategy Looks Like When You Can Only See the Output
You have an agency’s public work and nothing else: the feed they run, two years of posts. The strategy document is one you will never see, and may not exist.
The one claim in the pitch that cannot be checked
Ask about strategy and you get three things: content pillars, audience personas, a posting cadence. None can be falsified from outside. All three describe intentions rather than decisions, can be written after the work and still fit it, and take an afternoon to produce.
The work is different. A deck was made for you, this week, in the room. A feed was made for someone else and cannot be revised. That asymmetry is the method, and it assumes the standard in what separates good social media work from average.
Define strategy, then, so it can be tested from outside: a consistent choice about who is addressed, formats that recur for a reason, posts that advance one argument. Without that, there is no evidence of a plan.
Strategy shows up as exclusion, not as coverage
Work with a plan behind it is visibly not for everybody. A feed built for one narrow reader assumes vocabulary and refuses to explain it: a term of art in a caption with no gloss, a joke that needs prior context to land, background skipped because the intended reader already has it.
A feed with no decision does both at once, defining a basic term in one post and assuming deep familiarity in the next. That is not range, it is an absence of choice.
Take five posts at random and ask who each one is unusable for. If nothing is unusable for anyone, no choice has been made.
The shuffle test
Here is my own reading method, not an established framework. Take the last twenty posts, hide the dates, shuffle them, and ask whether anything is lost.
Work that fails is self contained throughout: each post sets its own premise, makes its point, closes. Shuffled, nothing breaks, because nothing depended on what came before.
Work that passes breaks. A post halfway through answers an objection, and it lands only because a claim three weeks earlier set that objection up. Shuffled, it answers nothing.
Two things create that order: a position built on later, and a format that changes as the account learns. This reads only on work that actually shipped, since concept and campaign work was made in one pass.
Formats that repeat because they work, and formats that repeat because they are cheap
Repetition is evidence both ways. These signals separate them.
| Check | Repeats because it works | Repeats because it is cheap |
|---|---|---|
| What it carries | A different argument each time | The same argument restated |
| Effort over months | Holds or rises | Decays: shorter, flatter, less care |
| Revision | Structure has changed once | Identical to the first one |
| Fit | Suits what is being said | Cheapest thing to produce |
A talking head explainer works when the speaker knows something you cannot get elsewhere, and is filler when a script is read to a lens at habitual length. A text card series works when the cards are steps in a necessary order, and fails when a paragraph was chopped into four.
Two years of a format with no revision is a habit, not a decision.
Posts that belong to the same argument, not the same folder
Across twenty posts, look for a claim made early and returned to with new evidence, a position held where hedging was safer, callbacks that assume you saw the earlier post, and escalation from a simple idea to a harder one.
Thematic grouping looks similar at a glance: posts share a subject and a template, and never build. Shared topic is filing, not argument.
One honest complication. Some accounts are standalone by design, because the audience arrives cold. Tell that decision from the absence of one by whether it holds across every post, not most, and whether the format suits it.
What competent work with no strategy behind it looks like
The most common case is not bad work.
- Every post is individually defensible.
- The calendar is full and regular.
- Quality is even across months and never spikes.
- Nothing is risky. No post could have irritated anyone.
- The whole body could be reordered, or split between two unrelated brands, unnoticed.
Competence is not strategy. A full calendar is the easiest thing in this business to buy. The work may be worth paying for, but not for the reason it is being sold.
The limits of reading strategy from the outside
Three limits, before anyone over applies this. A plan can exist and have been overruled, so you may be reading a client’s decisions. The public work is a selected sample, and the agency often does not control what stays up. And none of it tells you whether the work did anything, which is a separate question.
A strategy you can see is not automatically a good one. It is a legible one. Legible beats unfalsifiable, but a clear plan can still be aimed at the wrong reader. Read the work for whether a decision was made. Whether it was right is judged separately.
Try it on work already read this way
Every selection on this site is editorial, nominated or scouted, with the reasoning written down. Run the shuffle test on the accounts behind them. See this year’s selections, or nominate an agency.
FAQ
Can an agency have a real strategy that is invisible in the work?
Yes. A client can veto the sharp version, a budget cut can collapse a plan into whatever is cheapest, and legal review can flatten a point of view. An invisible strategy is still not evidence of one: treat it as unproven, not absent.
How many posts do I need to look at?
Around twenty consecutive ones. Fewer and a single campaign dominates the sample. More and you are reading two eras of the account at once. Check the oldest posts separately to see whether anything changed.
Does a consistent visual identity count as strategy?
No. A template is a production decision, handed to anyone and executed without a thought about who is addressed. It becomes evidence when the visual choice excludes part of the audience on purpose, or changes when the argument does.
What if the agency only shows campaign work and no ongoing account?
A campaign was made in one pass, so it looks coherent by construction and cannot show the continuity this method tests. Read an account they run instead.
