Invisible KeyboardSignal ReportLast 90 days
LinkedIn activity intelligence

How OpenAI shows up on LinkedIn

Which OpenAI people actually post, what they say, and — now — what actually performs. The accounts, cadence, formats, and content worth building around.

46 employees screened13 active posters23 postsJun 6 – Sep 4, 2026
01

The headline

Across 46 OpenAI employees, only 13 posted anything original in 90 days — and three of them wrote more than half of it. The field is wide open, and the content that wins is more personal than you'd guess.

28%
of screened employees posted at all in the window (13 of 46)
1post
what the median active poster published across the entire 90 days
52%
of all posts came from just three people (Flavia, Samantha, Abhi)
4×
more engagement on text posts than video, on average
The single best post in 90 days wasn't a product launch — it was Felipe Ribeiro's plain-text note that he was joining OpenAI, which pulled ~1,300 reactions and comments, roughly 10× the median. Personal beats promotional, and it isn't close.
02

90 days, one lane per person

Every post as a dot on the 90-day window — one lane per poster, colored by format, and sized by the engagement it earned. Two voices carry the volume, most others surface once, and the biggest dots are personal posts. Hover any dot for detail; times are each poster's local timezone.

VideoImageTextsmall → large = more engagement

Timezones: BRT = Brazil, ET = US Eastern, PT = US Pacific. PT* = US poster with no city listed (Pacific assumed).

03

What actually performed

The 10 highest-engagement posts (reactions + comments + reposts). Personal career news and mission posts top the list; product announcements land mid-pack.

Average engagement by format

Plain text out-pulled produced video roughly four to one. Simple beats slick here.

Text
195
Image
146
Video
48

The takeaway

Employees here get modest reach — tens to low-hundreds of engagements, not the thousands the OpenAI brand page pulls. The exceptions are personal: a career move, a mission post, a point of view. That's the ceiling worth chasing, and it's reachable with words alone.

04

Anatomy of a post that works

The top performers share a shape. This is the template we'd write to — pulled straight from what already worked inside OpenAI.

Step 1The hook is personal news

First line states the news, no throat-clearing. No "I'm excited to announce" — just the thing.

"Some news. After eight years at Wieden + Kennedy, I'm joining OpenAI as Head of Creative for Latin America."Felipe Ribeiro · 1,316 engagements

Step 2Tell the arc, not the résumé

Where I came from → the leap → what it means. A short story, not a bullet list of responsibilities.

"Career update — I've joined OpenAI as a PM on the Codex team! Over the last few years I've seen the incredible progress… Codex has become the primary way I do almost all of my work."Neil Garde · 482 engagements

Step 3Say the hard part out loud

Vulnerability and mission out-earn polish. Naming the tradeoffs is what pulls comments.

"Working on OpenAI user safety is not easy. It comes with privilege, hard tradeoffs… it has also made work feel more meaningful."Jesika H. · 236 engagements

Step 4End on a soft ask

A light CTA — hiring, "come build with us," or a question — turns readers into commenters without feeling like an ad.

"…there are few problems more consequential today than helping make AGI safe. Come work with us!"Jesika H.

Step 5Keep the format plain

The best posts were text or one simple image. Don't over-produce — a written story out-performs a polished video here 4×. Save video for demos, not moments.

05

The patterns

Across all 23 posts — directional, not statistical, but a consistent shape for planning a calendar.

Format mix

Text is both the most-used and the best-performing format.

Text
13
Image
6
Video
4

Time of day

Mid-morning to early afternoon owns it — 57% land between 9am and 3pm local.

12–6 am
2
6–9 am
2
9–12
6
12–3 pm
7
3–6 pm
3
6–9 pm
2
9–12 pm
1

Day of week

Tuesday leads; a soft weekday spread with a few Sunday posts.

Mon
4
Tue
5
Wed
4
Thu
3
Fri
3
Sat
1
Sun
3

Who & where

SF-heavy, with Brazil a strong second — two Brazil-based voices carry it. Counts are posts.

SF / Bay
12
Brazil
7
US (other)
3
New York
1

Gender of posters

Women · 13 posts
Men · 10 posts
6 women & 7 men among the 13 active posters — women post more, driven by Flavia and Samantha. Gender inferred from profiles.
06

What we'd do

Six plays, now backed by what actually performed.

PLAY 01

Own the "I joined" moment

The single highest-performing post type is the personal joining-OpenAI announcement — 10× the median. Every new hire has one to give. We'd make sure none of them gets wasted on "excited to announce."

PLAY 02

Activate the quiet majority

Nearly three-quarters of the people we screened posted nothing in 90 days. We'd pick 8–12 high-signal employees and get them posting consistently with a done-for-them system.

PLAY 03

Lead with the human layer

Personal and mission posts out-engaged product announcements decisively. We'd build a signature narrative series per person — career arcs, "how I use it," the hard parts — not a feed of launches.

PLAY 04

Write plain, not produced

Text averaged 4× the engagement of video. We'd put the effort into the first line and the story, not the edit — and save video for genuine product demos.

PLAY 05

Own the LATAM & multilingual lane

Two Brazil-based voices, posting in Portuguese, drove a third of all posts. Regional, multilingual employee content is wide open — a natural fit for our EN / ES / PT range.

PLAY 06

Ship Tue–Thu, mid-day

The reliable window is 9am–3pm local, Tuesday through Thursday. We'd program a steady weekly rhythm there — and test the quiet Sunday slot — so presence is planned, not sporadic.

Invisible Keyboard

Your team's best posts are still unwritten.

You already have the voices, the stories, and the proof that this works. What you need is the system. We build LinkedIn presence for founders and executives at AI and B2B tech companies — strategy, ghostwriting, employee content, and ICP outbound in English, Spanish, and Portuguese. So your people show up consistently without it becoming another job.

Make content people trust.
07

How this was built & what to trust

Method & honest caveats

  • Source: LinkedIn People search filtered to current company = OpenAI (six result pages), then each person's public activity feed. Post dates and times are decoded from each post's LinkedIn ID.
  • Sample, not a census: 46 screened employees. LinkedIn orders results by relevance to our network, so the sample skews toward SF, Product, and our connections.
  • Engagement: reactions + comments + reposts each post received. Comments and reposts are exact; reaction counts are read from LinkedIn's summary and are approximate on a few posts.
  • Window: Jun 6 – Sep 4, 2026 (last 90 days).
  • Inferred fields: gender is inferred from names/profiles, not self-reported. Times use each poster's local timezone; where a city wasn't listed, Pacific was assumed (PT*).
  • Sample size: 23 posts is small — read patterns as signals to test, not settled facts.
Invisible Keyboard · Make content people trust
invisiblekeyboardai.com · daniela@invisiblekeyboardai.com · LinkedIn signal report, Sep 2026 · built from public LinkedIn activity