The screenshot beats the infographic

What did we find when we scraped our own LinkedIn account?

We scraped four LinkedIn accounts this quarter, ours and the three accounts closest to ours in audience and subject matter, 342 posts in total and between 71 and 99 per account. For each account we measured reactions per post, and we ranked the four feeds on production value, meaning the visible cost of making a post look finished: designed templates, branded color, careful typography.

Our account ranked first of the four on production value and last on median reactions. We measured a median of 24 reactions per post on our feed against medians of 88, 100 and 189 for the three peers, about a quarter of the median of our nearest peer.

What the measurement shows is a rank order across four accounts: highest production value next to lowest median reactions. It does not show what caused the gap. Format is the candidate explanation this post pursues, because the three accounts with higher medians share a posting format we almost never used, and our default, the designed infographic, appears in none of the five top posts on the peer with the highest median. Audience size and the poster's standing are the two explanations we did not control for, and four accounts cannot separate them from format. The test that can, one account, two formats, the same period, is described at the end.

definitionPosting the screen

Posting the screen is publishing a capture of real work instead of a designed summary of it. The dashboard as it rendered, the sketch as it was drawn, the tool mid-run with the output still loading. It travels because it is evidence. A reader can see that the work exists, and evidence is the one thing a feed full of compositions cannot supply.

Why would a polished infographic lose to a screenshot?

Two mechanisms would explain the gap if format is the cause, and the scrape tests neither directly. First, the reader of a B2B feed wants to know whether the author does the work being described. A live dashboard, a working folder or a tool mid-run is only in that state when somebody built the thing and ran it, so a screenshot answers the question; an infographic answers whether the author has access to design, which is easy to buy. Second, an infographic arrives finished, and the reader's only move is to accept it or scroll past. A screenshot arrives raw and asks for a small act of verification, reading the numbers, noticing the tab names, deciding what the layout implies, and a reaction is one trace that act can leave.

We did not choose wrong on purpose, and neither did the teams whose feeds look like ours did. Polish is the correct standard in most other channels a marketing team ships to. The feed selects for something else, that standard is rarely written down, and reaction counts are public while the counterfactual is invisible, so a polished feed reads as a modest success rather than a measured loss.

What do the winning posts have in common?

Three different formats with one shared property: each is proof of a person doing real work, captured rather than staged. Here is the structure of one.

The top post on one peer account in our scrape, the dashboard screenshot, opens with a single operational fact: who runs the spend, how much per month, and from where. It is about 240 words in 26 short lines, 12 of them an inventory, one tool per line with a one-clause description. One claim, that the work runs from a terminal, backed by one effort figure and one count. It ends with three plain sentences on how to use the tools and a download link.

The format all three peers use heavily and we almost never used is the screenshot of real tooling doing real work. Our delivery work produces those screens daily: ad batches mid-generation, scoring passes, reject galleries with the rule that killed each item, working folders with the file names showing. We had been converting that evidence into compositions, and if the format explanation holds, the conversion was subtracting the value.

Do AI engines reward the same format the feed rewards?

The overlap is measurable, because we ran a separate audit of how AI engines answer the buying questions in our market. Across the 703 citations that audit collected, counting each cited domain once per answer, linkedin.com was the single most cited domain, ahead of every publisher, analyst and vendor site, with Reddit second. The cited LinkedIn items were mostly individual posts by individual operators rather than company pages, and every one of them was cited by Perplexity.

The cited posts share a shape that can be built on purpose.

The count is in the 37 cited posts, read from their first lines.

The same audit measured which questions earn citations at all. ChatGPT cited sources on comparison questions 75% of the time in our run and did not cite definition content once. Perplexity cited sources on all 36 questions. Claude leaned the other way and cited definitions 67% of the time. Comparison was the only band with a meaningful citation rate in all three engines at once, so the highest-value post takes a position on a genuine either-or question a buyer is already asking, the shape of should we hire a GTM engineer or work with an agency. The engines cite operators by name, so the citable asset accrues to a person willing to hold a position in public, and a company that wants the citation has to let a human be the author.

What should a B2B team change about its LinkedIn content?

Post the screen before you pay for the polish, and measure the two against each other, which is the test we describe at the end. In practice that is five changes.

• Publish captures, not compositions: the dashboard, the working folder, the reject pile, the tool mid-run. When a post needs a design pass, ask what the pass is covering for.
• Write as the operator, in the first person. Both the feed and the engines select for a person doing the work over a page describing it.
• Make one claim per post and attach a number or a named mechanism to it.
• Take a position. Comparison questions were the only band cited by all three engines in our run, and a comparison answer needs a stance.
• Stop paying for production value until you have tested it. Ours was the highest of the four accounts we scraped, next to a median of 24 reactions, and the test below is how we find out whether the two are connected.

One constraint is real. A screenshot of client work is client data, and it needs written permission or it stays private. The workaround is the work you run on your own systems, which most operating teams have more of than they publish.

We wrote our kill criterion before changing anything, because a firm that recommends measurement should be seen submitting to it. Four accounts cannot separate format from audience size or standing; a same-account test can, one account, two formats, the same period. If posting the screen does not move our median reactions within six weeks, the format explanation fails and we will publish that result with the same numbers attached. That is the standard we hold any go-to-market claim to in how we work.

The infographic era of our feed produced the most professional-looking account in its peer set and about a quarter of the median reactions of its nearest neighbor. A scrape we ran against ourselves, and lost, is what changed our behavior. Run the same scrape on your account against its three closest peers. The result will be uncomfortable in one direction or the other, and either way you will know what to post on Monday.