
The Stakes
August 2026
Three in four people say chief executives are obligated to help bridge the trust divides inside their own organizations. Only forty-four percent are judged to be doing it well. That is the 2026 Edelman Trust Barometer, and the shortfall is not a shortfall of effort.
Most of those executives are communicating more than they ever have, and more easily than they ever have.
Output has never been cheaper. Belief has not moved with it.
Which makes the question for this issue narrower than it first looks.
When everyone in your market can produce the words, what is left that a reader can actually check?
Welcome to Issue 8 of Narrative Yield.
Last issue closed on ownership: if the distribution you rely on disappeared next quarter, what would still be yours? This one narrows the question. Say the position is yours. Can anyone tell, from reading you, that it is?
I went through SAP's presales academy in my twenties, which is where I learned to run a clean demo. Everyone trained in that room could deliver the same polished thirty minutes, and most of us did. The meetings that actually moved were the ones where somebody said a thing only they could have said, adding context and building off the script we had all been given.
I have thought about that a lot this year. The polished thirty minutes are now free. The part they added is the only piece left with a price on it.
The Signal
BEN HORWITZ

Ben Horwitz, Sinceerly, now running Startups @OpenAI
In April, Business Insider reported on a Chrome extension called Sinceerly. It does one thing: It puts the typos back into your email.
Ben Horwitz built it with Anthropic's Claude, over about a month, between classes. His own description of the project was "I made the anti-Grammarly."
The premise is that flawless spelling and grammar now reads as a machine. A dropped space, a doubled letter, a lowercase i: those read as a person typing quickly, which is to say someone real. Katie Notopoulos, reporting it, made the larger observation that typos have crossed over into a status marker. Garbled text signals someone senior enough not to proofread, and busy enough to be worth reading.
Then Horwitz tested it. He cold-emailed five Fortune 500 chief executives using the extension.
Four of them replied. Not one reply ran longer than ten words. Two of them contained typos. One executive addressed him as Larry.
Sit with that for a minute. A business school student wrote a program, using AI, that fabricates the appearance of human haste. It cleared a filter that four Fortune 500 chief executives were applying at the top of their inbox, whether or not they could have named it. Then the replies came back short, misspelled, and addressed to the wrong man.
The extension ships a preset called CEO mode: all lowercase, cut to the bone, "sent from my iPhone" appended whether or not you have a signature. Executive roughness is now a dropdown.
The signal worked. It cost a month of evenings and an API key.
Whatever roughness was worth as evidence before April, that is roughly what it is worth now.
The Substance
There is a harder version, and it ran three months later.
On July 30, the US State Department presented a map of Africa at the AIDS 2026 conference in Rio de Janeiro. Every country on it was mislabeled. Nigeria, which has a coastline, sat in the Sahara. Ivory Coast was landlocked. A Reuters analysis found an OpenAI watermark in the file. The Department apologized the next day, explaining that a team member had "hastily altered the slide deck immediately before the presentation."
Screenshots moved through policy networks within hours, and Semafor's reporting caught what made it land: officials read the error as a signal about institutional capacity. The map was fluent, professional, and instantly produced. What it disclosed was that nobody in the room knew the continent well enough to catch it.
So both signals are spent. Polish stopped meaning care the moment a model could generate it on request. Roughness replaced it and lasted about a year before someone shipped it as an extension. Both rode on the surface of the artifact, and the surface is the layer that can now be manufactured.

Which leaves the layer underneath. Two memos, three weeks apart.
On April 7, 2025, Shopify's Tobi Lütke published a memo saying reflexive AI usage was now a baseline expectation, that AI questions would enter performance review, and that any team requesting headcount would first have to explain why the work could not be done by AI. On April 28, Duolingo's Luis von Ahn published his own: Duolingo would be AI-first, would gradually stop using contractors for work AI could handle, and would add headcount only where a team could not automate further.
Both memos say the same thing where it counts. Prove the machine cannot do it before you ask for a person.
The reception was not the same. Duolingo posted to LinkedIn and drew over a thousand comments, hundreds of reposts, and a run of users announcing they were deleting the app. The company wiped its social accounts and lost more than four hundred thousand TikTok followers in weeks. Von Ahn walked it back inside a week, again that August, and again this April, when he told Fortune he would not require employees to be evaluated on AI usage at all.
Two fair objections, worth stating rather than stepping around. Duolingo is a consumer brand with a mascot and an app-store audience. Shopify sells to merchants, and merchants do not leave one-star reviews. And von Ahn named contractors explicitly, where Lütke did not. Both are real, and neither is the size of the gap. The texts differ by a clause. The outcomes differ by a brand crisis.
What differed most was what the market already had on file. Lütke had been writing publicly for years about how Shopify builds things, so the memo landed on top of a record. Duolingo's most visible public voice was its mascot. The memo arrived with no prior, so it became the whole exhibit.
That is what receipts look like: the one signal you cannot install in a month, a record of specific, dated, checkable claims under your own name, held long enough that a reader can see whether you were right.
The practical version is smaller than it sounds. Take the last five things you published. Count how many contain something that could be shown to be wrong: a number, a prediction with a date, a position someone in your market would argue with. If the honest answer is none, you have been producing polish, and your market has no way to tell it from the machine.
The Shift
Issue 7 asked what would still be yours if the distribution disappeared. This one asks whether what is yours is legible as yours.
Somewhere in your market there is someone with less capability who has been publishing checkable claims under their own name for three years. When a buyer, a board, or a reporter goes looking for a read on your category, that person has a record and you have a presence. The record wins, because a record can be checked and a presence cannot.
You cannot buy that, and you cannot install it. That is the good news, because it means the only people holding it are the people who did the work.
Polish is free. Roughness costs a month. A record costs years, which is precisely why it still means anything.
So say something this week that could be proved wrong.
P.S. If you want an honest read on where your own signal stands today, the Executive Visibility Scorecard takes about five minutes.
Until next time,
Michaella
Narrative Yield is a newsletter about the stories that drive revenue, investment, and trust. If this landed for you, forward it to the leader in your life who has the substance but hasn't found the story yet.
Did a friend forward this to you? You can subscribe here
Sources
2026 Edelman Trust Barometer (75% say CEOs are obligated to help bridge trust divides; 44% judged to do it well): https://www.edelman.com/trust/2026/trust-barometer
Ragan, "By the numbers: Edelman Trust Barometer 2026": https://www.ragan.com/by-the-numbers-edelman-trust-barometer-2026/
Business Insider, Katie Notopoulos, on the Sinceerly Chrome extension (April 2026): https://www.businessinsider.com/ai-written-email-perfect-typos-new-chrome-plugin-2026-4
Futurism, "New Browser Plugin Adds Typos to Your AI-Generated Emails to Make Them Look Real": https://futurism.com/artificial-intelligence/browser-plugin-typos-ai-generated-emails
CNN, "State Department says it made 'an unfortunate error' when it mislabeled countries on map of Africa" (July 30, 2026): https://www.cnn.com/2026/07/30/politics/state-department-africa-map-mislabel
NBC News, "State Department map of Africa mislabels every country at AIDS conference": https://www.nbcnews.com/politics/trump-administration/us-government-map-africa-mislabels-every-country-global-conference-rcna590153
Washington Post, "State Dept. apologizes for mislabeling countries on Africa map" (July 30, 2026): https://www.washingtonpost.com/world/2026/07/30/state-dept-apologizing-africa-map-that-mislabeled-every-country/
Futurism, "US Government Displays AI-Mangled Map of Africa That Gets Every Single Country Wrong" (OpenAI watermark, per Reuters analysis): https://futurism.com/future-society/us-officials-ai-map-africa
TechCrunch, "Shopify CEO tells teams to consider using AI before growing headcount" (April 7, 2025): https://techcrunch.com/2025/04/07/shopify-ceo-tells-teams-to-consider-using-ai-before-growing-headcount/
CNBC, "Shopify CEO says staffers need to prove jobs can't be done by AI before asking for more headcount" (April 7, 2025): https://www.cnbc.com/2025/04/07/shopify-ceo-prove-ai-cant-do-jobs-before-asking-for-more-headcount.html
PR Daily, "The Scoop: Duolingo CEO walks back 'AI-first' memo": https://www.prdaily.com/the-scoop-duolingo-ceo-walks-back-ai-first-memo/
Entrepreneur, "'Continuing to Hire': Duolingo's CEO Clarifies AI Stance After Backlash": https://www.entrepreneur.com/business-news/duolingo-ceo-clarifies-ai-stance-after-backlash-read-memo/492141
Fortune, "'I'm not going to force you': Duolingo CEO backs off from evaluating employees on their AI usage" (April 13, 2026): https://fortune.com/2026/04/13/duolingo-ceo-luis-von-ahn-ai-usage-requirement-employee-performance-evaluations/
