How to Stop Spinach Posts Sounding Like Everyone Else’s
Spinach posts sound generic for a structural reason: the model summarises what your sources say, and most company docs describe rather than argue. Fix the source material and the output changes within a week, without touching the prompt.
The three changes below took my test account from “could be any SaaS” to recognisably one company’s voice.
Why does the output sound generic?
Because “we launched a feature that helps teams do X faster” is the shape of every product update ever written. If the input contains no contrarian claim, no specific number and no named person, the output cannot contain them either.
Most B2B internal documents are written to be safe. A safe input produces a safe post, and safe posts are invisible.
Fix 1: add contradiction, not volume
One document where someone disagrees with the industry consensus does more for output quality than ten additional transcripts. Connect a source that contains an opinion: a founder’s market memo, a sales call where a prospect pushed back, a competitor teardown.
In my test, adding a two-page memo arguing against a common metric changed the ratio of posts with a clear claim from about 1 in 10 to 6 in 10.
The same principle applies to the voice settings in Spinach: negatives work better than adjectives. List phrases the brand never uses.
Fix 2: give it a number to carry
Posts with a specific figure read three times as substantial, and figures rarely appear unless they are in the source. Keep a running stats doc, updated weekly: churn percentage, average close time, tickets solved, whatever you actually measure.
Then seed the queue. Most generators allow a topic or seed input per post; a seed of “support tickets down 22% after the new routing” produces a concrete post, while a blank seed produces a summary.
Caution: numbers need context. “22% reduction” without a baseline gets questioned in replies. Write the baseline in the stats doc so the post can include it.
Fix 3: cut the template phrases explicitly
Add a banned list to the brand voice and revisit it monthly. The phrases that recur most in generated B2B posts:
- “We’re excited to announce”
- “Take your X to the next level”
- “In today’s landscape”
- “Here’s why that matters” as a standalone line
- “Let’s dive in”
Removing these forces the model into plainer constructions. That alone reads less templated, even when the underlying content is thin.
What still needs a human
The opening line. Generated openers default to a summary of the post, which is the least engaging option. Rewrite the first sentence of every post you approve — it takes 15 seconds and changes the result more than editing the body.
LinkedIn versus X. Same source, similar text. On LinkedIn, a longer opening context works; on X, the first six words decide everything. Change the platform of at least one post per day manually rather than accepting the default.
Images. Generic. Replace every hero image with a screenshot, a chart or a real photo. This is the single highest-effort item in the loop and the one most worth doing.
A one-hour weekly routine
- Update the stats doc with three new numbers.
- Add one opinion source — a memo, a customer call, a competitor note.
- Review the queue top to bottom, rejecting anything that could belong to another company.
- Rewrite the opening line on every accepted post.
- Replace images on the three posts with the widest reach.
That is about 45 minutes for a week of 7 to 10 posts.
Who should not bother
If your team has no written opinions anywhere — no memos, no pushback in support threads, no sales calls with objections — this tool will produce wallpaper. Spend a month writing weekly internal notes first; the content itself is the asset.
Solo creators also struggle, because the volume of source material is too low for the model to find anything specific.
Quick Summary
- Generic output is usually a source problem, not a prompt problem.
- One contrarian document beats ten neutral ones; it raised claim-bearing posts from 1 in 10 to 6 in 10 in testing.
- Keep a weekly stats doc and seed every post with a specific figure.
- Ban template phrases (“we’re excited to announce”, “take your X to the next level”) in the brand voice.
- Rewrite the opening line and swap the image on every approved post — highest return per minute.
