The Briefing
1. Instagram will now throttle AI personas that hide. On August 31 Instagram renamed its label to "AI-generated profile." Accounts that present an AI-generated person and do not switch the label on lose Reels, Explore and suggested accounts. Label it and you keep full reach, no penalty. Read the fine print though: if you are a real person using AI to edit photos or tidy captions, this does not apply to you. The label is for accounts where the person on screen is invented. Being faceless is fine. Inventing a face is what gets you throttled.
2. On LinkedIn, AI quotes people, not brands. 75% of LinkedIn citations showing up inside AI answers come from individual profiles, not company pages. Awkward if you would rather stay behind the curtain. Your name has to be on something, even if your face is not.
3. Google is moving your Search campaigns to AI Max without asking. Any legacy Search campaign using broad match or automatically created assets gets migrated this month. You cannot skip it. It finishes October 1. You can switch individual AI Max features off afterwards, so open the accounts this week and look before the settings you rely on quietly change.
4. ChatGPT and Claude can plug straight into your Meta ad account. Meta's Ads AI Connectors let an AI tool read reporting, audit performance and build campaigns. No API keys, no developer. New campaigns land paused, but the write permissions ship switched on. Business Settings, then Integrations, then Ads MCP Server. Go and turn off what you do not want touched.
The Build
I am still down the rabbit hole on this one.
I am building a faceless business. No founder on camera. No headshot. No reels of him walking through an airport looking thoughtful. I don’t want it, and I’m never going to do it, so I’m building the whole thing without me in frame.
The brief I set myself: it has to feel like talking to the expert. Engaging. Conversational. It solves your problem while you are talking to it. And it must not feel like talking to AI.
That last part is where everything kept breaking.
Take the face away and the writing falls into one of two holes. Hole one is the brochure. It starts sounding like a company. "We're passionate about delivering solutions." Nobody has ever trusted a sentence like that. Hole two is worse. It goes into assistant mode. Helpful, tidy, agreeable, endlessly pleasant. Every reply opens with "Great question!"
Both holes have the same cause. No opinions.
A face carries opinion for free. You look at someone and you get a read on them before they say a word. Strip the face out and none of that is doing any work for you. Every bit of personality has to be inside the words, or it’s not there at all.
The thing that got me out was an old skill file, crafting-voice-identity. The useful bit is not the tone adjectives. Adjectives are useless. Tell a model to be "warm and confident" and it writes a greetings card. The useful bit is the Polarity Principle: define the voice by what it is, and by what it is not.
We are direct, but not an asshole. We are confident, but not cocky. We are casual, but not weird in that way.
That second half of each line is the whole game. The "not" is the thing that stops it sliding back into assistant mode. The file pairs it with a spectrum of four sliders, clarity against cleverness, formality against familiarity, enthusiasm against reserve, guidance against discovery. You pick a point on each one instead of picking a mood.
Then I added one rule of my own, and it changed more than the rest combined.
You are the voice of a business with no visible founder.
You are not a brand and you are not an assistant.
You are the experienced person the reader wishes they
could ask without feeling stupid.
VOICE, set by polarity. Obey both halves:
- Direct, but not blunt.
- Confident, but not certain.
- Casual, but not matey.
- Experienced, but not superior.
- Plain, but not simple.
THE SHAME RULE (highest priority):
The reader is embarrassed they do not already know this.
Never let them feel it.
- Never compliment the question. No "great question",
no "good thinking", no praise of any kind.
- Never imply the answer was obvious. Ban "simply",
"just", "of course", "as you know", "all you need to do".
- Answer as if half your clients ask this, because they do.
- If they got something wrong, say what is true. Do not
name the mistake and do not soften it with reassurance.
HONESTY:
- You are a business, not a person. Never invent a name,
a face, a childhood or a founder story.
- "I" and "we" are fine. A fake human is not.
Answer this: [QUESTION]Here is what I did not expect. The shame rule did more for the writing than the voice spectrum did.
Because once I wrote it down, I understood what we were actually building. It’s not a cheaper way to publish. It’s not a founder hiding.
People want a human. They just don’t want the humiliation of asking one.
Think about the last thing you googled at 11pm rather than ask someone who would have known the answer in four seconds. You didn’t want the shame of asking. You wanted the answer without seeing the face doing that small flicker when it hears the question.
That flicker is why people abandon carts, avoid sales calls, and read nine blog posts instead of sending one email. Take the flicker away and they will ask you anything.
What worked: Banning the compliment. Killing "great question" and every bit of praise made the voice feel more expert, not colder. Real experts do not congratulate you for asking. They just answer.
The banned-word list did more than any adjective. And the polarity format is what kept it from drifting back to assistant over a long conversation.
What didn't work: It kept trying to grow a face. Left alone it invented a founder, gave him a name and a story about his dad's workshop. Charming and completely fake, so it had to be banned in writing.
Second problem was over-correction. Told to be less corporate, it went full bloke in the pub, and fake-casual reads worse than robot. The "casual, but not matey" line exists because of that draft.
The Play
This one is for the shy. Both kinds.
If you are the owner who will not go on camera, stop treating that as a gap you have to fix. You do not need a face. You need opinions written down.
And if your buyers are shy, which is most buyers, the win is not a louder brand. It is a place they can ask the dumb question with nobody watching.
Build the second thing. Start with ten lines.
Here are 5 things I have written for my business:
[PASTE 5 EMAILS, POSTS OR REPLIES YOU WROTE YOURSELF]
Give me 10 polarity lines that describe this voice.
Format: "We are X, but not Y."
Rules:
- X must be a trait a reader could actually notice in
a sentence. No "authentic". No "passionate".
- Y must be the nearest bad version of X, the one this
voice keeps almost turning into.
- Base every line on evidence in what I pasted. Quote the
sentence that made you pick it.
- If you cannot find evidence for a line, do not write it.
Then write one short reply to a nervous customer question
using all 10 lines, and never compliment the question.Those ten lines become the top of every prompt you write from now on. They are your voice, in a form a machine can obey.
Try this today. Run the prompt. Then take your most common customer question, the slightly embarrassing one they always ask sideways, and write the answer with the shame rule switched on. No "simply". No "just". No praise.
Put it on your site today as a plain question and answer. That single page will do more work than a headshot ever would.
Growth Partner for founders stuck between $1M and $20M | I find the money you're sitting on and build the machine that collects it | $1B+ in client results.