Every brand remembers how it sounds.
Brand Memory holds each brand’s voice, audiences, products, knowledge and style rules. Every draft and every agent reads it, and it learns from what your posts actually did.
Personal, punchy and data-led, with a candid streak. One builder sharing what they shipped and what it did.
- Lead with a number, and only real ones.
- Write in the first person.
- Short lines, one idea each.
- A link in the first part.
- Hedging: “might”, “kind of”.
Pick posts that sound like you.
A voice guide from a blank page is slow. Tick a few sample posts that sound like the brand, two or three is enough, and Harbor proposes a voice: personality, values, do and avoid rules. Or paste your own posts, or import them from a connected account.
- Six sample personalities, two posts each
- A strength meter shows when there’s enough to learn from
- Nothing is saved until you choose “Use this voice”
Dial the voice in, and read the result.
Five dials, each between two poles: company or personal, calm or punchy, story-led or data-led, serious or playful, polished or candid. A sample post on a fixed topic rewrites as you drag, so you hear the change before saving. Push a dial past a third of the way and it adds a plain rule.
- The dials start from the posts you picked
- Rules a dial now contradicts are dropped
- Go on, drag one: it’s the app’s own logic
I just shipped something new. It does one thing, and it does it well. Try it and let me know what you think.
Rules with receipts.
The voice refresh compares the brand’s posts that beat its median with the ones that didn’t, over the last 90 days. When something like opening with a number lifts results by 1.25× or more, with at least 3 posts on each side, it proposes a do rule. At 0.8× or below, an avoid rule.
Rules your results contradict are proposed for removal. You approve changes the same way you approve posts.
- Every change shows its lift and the posts behind it
- Tick the ones to keep; unticked changes are rejected
- Run it by hand or on a schedule, once 6 posts have a verdict
- + AddVoice · Do↑ 1.6× medianOpen with a number.Posts opening with a number did 1.6× the median (7 vs 11 posts).▸ Evidence · 3 posts3 weeks, 2 rewrites, 1 feature. It does one thing…↑ 2.4×4 minutes to set up a brand, down from 15. Here’s what we cut…↑ 1.8×6 brands, 1 inbox: how I keep the voices apart↑ 1.4×
- + AddVoice · Avoid↓ 0.6× medianA link in the first part; put it in a reply.Posts with a link up top did 0.6× the median (4 vs 14 posts).
- − RemoveVoice · Do↓ 0.8× medianEnd with a question.Posts ending with a question did 0.8× the median (6 vs 12 posts).
- + AddVoice · Excerpt3 weeks, 2 rewrites, 1 feature.The best line of this month’s top post (2.4× the median).
See exactly what memory changes.
Write a one-line brief or pick an idea, choose a platform, and Harbor drafts it twice with the same model: once with the brand as it is, once with an empty memory. Words that only appear with memory are highlighted, and every rule that fired is listed with its evidence.
- Do and avoid rules, banned words, terms, products and audiences
- Each fired rule links to the posts that prove it
- Nothing is saved; it’s one recorded run
Keep the good stuff where you find it.
On any post, reply, reply theme, signal or draft, one action saves it to Brand Memory as a knowledge note, a verbatim excerpt, an example post, or a do or avoid rule. It stays linked to where it came from, so the rule carries its evidence.
- A reply that asks a good question becomes knowledge
- A post that nailed the tone becomes an example
- Duplicates are refused, so memory stays tidy
How is the growth curve measured? Views, or likes? Would love a post on this.
Plan next week in an afternoon.
Self-host Harbor free, or get early access to Harbor Cloud. Either way, nothing posts until you say so.