Founders are now publishing through voices that are not clearly people. AI-drafted essays, ghostwritten founder posts, support personas, and model-fed brand voice documents all create the same operating question.
What is this voice allowed to claim?
The question feels modern. It is not. In 1945, General Mills had already written a rule for an invented spokeswoman. The rule was simple, legalistic, and harder to satisfy than most AI content policies written today.
The rule
The April 1, 1945 Fortune profile of General Mills includes a line most modern summaries of Betty Crocker do not quote.
Anything said by Betty or credited to Betty must be literally true with respect to some current member of the company's home economics staff.
That is the standard.
Not plausible. Not on brand. Not likely to be true somewhere in the company. Literally true of a current employee.
This was not written for a real spokesperson. Betty Crocker did not exist as a single person. She was a corporate persona, created to answer household questions at scale, sell confidence, and make a milling company feel human.
The important fact is not that she was fictional. The important fact is that General Mills did not let the fiction float free. It tied the voice to living expertise inside the company.
The common story
The familiar version runs like this.
In 1921, the Washburn-Crosby Company, later part of General Mills, ran a Gold Medal Flour promotion in the Saturday Evening Post. Later company accounts describe it as a puzzle promotion that drew roughly 30,000 responses, with some carrying unrelated baking questions. A current General Mills history page says there were 30,000 completed puzzles and several hundred question letters.
The advertising department wanted to answer those letters, but Samuel Gale reportedly believed that a woman should sign the replies. The surname Crocker came from William G. Crocker, a company director. The first name Betty was chosen because it sounded friendly. The signature was selected from handwriting submitted by female employees.
That is the version you will find, with small variations, across institutional and brand accounts. The Minnesota Historical Society's MNopedia entry gives the broad outline and adds useful chronology, including the Betty Crocker Home Service Program radio premiere on October 2, 1924. General Mills' own pages vary on the volume of early questions. Its 2024 page says several hundred. Its 2022 profile of the women who built Betty Crocker says thousands.
The legend is not useless. It is just too smooth.
The corrections
The value of the 1945 Fortune profile is that it interrupts the polished story. It does not support three claims that are now repeated as if they were settled.
This is not a debunk for sport. The corrections matter because they make the surviving policy more interesting. General Mills did not build trust by having one imaginary woman personally care for every household. It built a managed system, then wrote rules for what that system could say.
Not thirty thousand
Fortune describes the original promotion differently. It calls it a recipe contest that drew "several thousand entries."
That is not the same as a jigsaw puzzle drawing 30,000 responses. The 30,000 figure appears in later General Mills retellings, not in the contemporaneous Fortune account cited here.
The distinction is not cosmetic. A puzzle story makes Betty sound like an accidental byproduct of mass engagement. A recipe contest makes her sound closer to what she became, a response mechanism for practical household expertise.
Not every letter
Fortune also cuts against the sentimental version of Betty as a personal correspondent to America.
By 1945, the mail room opened the "4,000 letters Betty averaged each working day." It answered about 80 percent with form letters. The remainder went to Marjorie Child Husted's department, which Fortune says had forty assistants, for individual attention.
That is still an extraordinary operation. It is not the same as every letter being personally answered.
The system worked because it separated common questions from exceptional ones. Most households got a tested response. Some got a human response. The persona sat on top of both.
Not a poll
The third correction is the most repeated.
Fortune opens by asserting that, "without much doubt," the woman best known to United States housewives, except Mrs. Roosevelt, was Betty Crocker. It does not publish a poll. It does not show a ranked list. It does not document the claim as a formal national popularity study.
Later headlines about Betty being the second most popular woman in America are paraphrases of an editorial sentence, not evidence of a public ranking.
The documented number comes later in the same Fortune piece. General Mills' own survey found that 91 percent of U.S. housewives knew of Betty Crocker, and 56 percent correctly connected her with General Mills. Those are company research figures, valuable but not independent.
The unquoted part
The part founders should care about is the written Betty Crocker policy.
Fortune says it was drawn up by the advertising and legal departments as a guide for copywriters. The policy made the boundaries explicit. Betty could be friendly, but not intimate. She would stick to home economics. She would not discuss a private life. She would say "we" rather than "I."
The reason was blunt.
It would be unsound and perhaps illegal to represent Betty as a superwoman who thinks up all her own recipes and answers all her own mail.
Then came the controlling clause.
Anything said by Betty or credited to Betty must be literally true with respect to some current member of the company's home economics staff.
This clause does three things at once.
- It admits the persona is not a person.
- It prevents the persona from inventing expertise.
- It binds every claim to a living employee who can substantiate it.
That is the durable lesson.
Betty Crocker was not treated as an all-knowing fictional woman. She was treated as a packaging format for real expertise. If the expertise did not exist inside the company, the persona was not allowed to speak.
Most AI content governance today is weaker. It asks whether the content was reviewed, whether the tone is consistent, or whether disclosure is needed. Those are useful questions. They are not the deepest question.
The deeper question is, who on the team is this literally true of?
What it bought
This is the section many marketing versions of the story skip.
Fortune reports that General Mills carried Betty Crocker on its balance sheet at one dollar, against roughly ten million dollars spent creating her. It also reports the company survey cited above, 91 percent awareness among U.S. housewives, 56 percent correct association with General Mills.
Those are meaningful signals. They suggest that the persona had become a large trust asset by 1945.
But we should not make the story cleaner than the sources allow.
We found no dated sales or market share figure tying the correspondence operation to flour sales. Fortune notes that per-capita U.S. flour consumption was flat at about 155 pounds through the period. That fact does not prove Betty failed. It does mean the usual causal claim is unproven.
If someone attaches a revenue number to the letter-answering operation without a source, they have added something the record does not show.
The honest claim is narrower, and stronger. General Mills built one of the best-known corporate voices in America, at scale, and put legal constraints around what that voice could claim.
The 2026 translation
For an early-stage software company, the parallel is not nostalgic. It is operational.
A founder may now publish a LinkedIn post drafted by a model, revised by a marketer, approved in Slack, and posted under the founder's name. A company may publish a technical guide written from sales calls, product notes, and AI-assisted drafts. A support persona may answer with a name that sounds human, while drawing from a knowledge base no single human fully owns.
These are normal systems now. The risk is not the use of systems. The risk is that no one has written the boundary.
The 1945 policy maps cleanly onto the decisions founders are making this quarter.
- If the post says "I have seen this pattern," which human has seen it?
- If the guide says "we recommend," who inside the company can defend the recommendation live?
- If the persona answers a customer, what is it forbidden to say because no one on the team knows it?
- If the named expert leaves, what happens to the archive built on that person's judgment?
- If a model drafts the page, who becomes accountable for the final claim?
The line between "I" and "we" matters. Betty's policy preferred "we" because "I" would falsely imply one woman was inventing recipes and answering all mail. Many founder-led companies now have the opposite problem. They publish corporate knowledge as if it were personal experience.
Both errors come from the same source. The voice is allowed to say more than the human behind it can truthfully support.
Can founders copy?
Yes. The rule is not complicated.
You do not need a fifty-page AI governance document to begin. You need a one-page persona policy that a stranger could audit.
The test is whether the sentence remains true after the content calendar, ghostwriter, model, and founder ego have all touched it.
A copyable policy
Use this as a starting point.
- Every published claim must be literally true of a named person currently on the team.
- The byline must name a human who will take the question if a reader asks for evidence.
- The company voice may say "we" when the knowledge belongs to the team, not to one person.
- The company voice may not say "I" unless the named human wrote, reviewed, and accepts the claim as personal.
- The voice may not invent experience, customer exposure, technical depth, or operational judgment it does not have.
- Anything the team cannot answer live does not get published as advice.
- AI may draft, summarize, or restructure, but a named human owns the final claim.
- When the named expert leaves, the archive tied to that expertise is reviewed, updated, reassigned, or retired.
That policy is intentionally plain. It is stricter than a disclosure note. It is more useful than a tone guide. It asks the only question that matters when a voice is not fully a person.
Is this true of someone real, here, now?
This is also the editorial standard Better applies in its own content work, real expertise first, voice second. The byline is not decoration. It is the accountability layer.
Why disclosure fails
MNopedia notes that visitors to the General Mills kitchens sometimes arrived expecting to meet Betty Crocker, then had to be consoled by tour guides when they learned she was not real.
That detail matters.
By 1945, Betty's constructed nature had been discussed in a mass-circulation magazine. The company was not operating in total secrecy. Yet people still felt deceived.
Disclosure and belief are not the same thing.
A company can disclose that a persona is synthetic, assisted, composite, or corporate, and some readers will still experience it as a person. That is the risk founders assume when they let something that is not a person speak in the company's name.
The answer is not to avoid all personas. Betty Crocker shows that a persona can be useful, durable, and governed. The answer is to constrain the persona before scale turns ambiguity into trust debt.
Takeaway
The Betty Crocker legend is less important than the rule buried inside the primary source.
A brand persona is not automatically a lie. But it becomes dangerous when it claims knowledge, experience, or judgment that no current human can stand behind.
General Mills' lawyers understood that in 1945. They wrote the constraint down.
Founders publishing AI-drafted content in 2026 should do the same.
Every claim made by the voice must be literally true of a named human currently on the team.
Start there. Then publish.