Engines, not experiments: personal brand
AI personal branding that works is a hybrid system, and we can show you where the seams go
Fully automated personal branding is easy to demo. Connect a model to a scheduler, feed it a topic list, and posts appear. Run it for three months and the failure modes appear too: invented details, a voice that drifts toward everyone else's, and replies that tell people a machine is answering.
We build the other kind. The model does the first, heavy part of each job; a person makes the calls that carry your name. This page explains where we draw that line, and lets you draw it yourself first.
Three more short videos on AI tools and personal branding work
The answer first
Answer: AI personal branding means using language models to do the volume work behind a person's public presence: research, transcription, drafts, repurposing, monitoring and reporting. It works when a human owns three things the model cannot: the facts about the person, the decision about what they stand for, and every interaction with another human. It fails when those three are automated too. We design systems with that split built in, and enforce it with checks rather than good intentions.
The second video on this page shows the pattern in a single task: let ChatGPT do the first 70% of the work quickly, then finish by hand. Our job is to turn that habit into a reliable system.
Sort the jobs before anyone builds anything
Twelve jobs that make up a personal brand. Put each one in a column, then compare your answer with ours. Disagreement is useful; it usually reveals a risk someone has not priced in.
Free tool
Which personal branding jobs should a machine do?
Sort twelve real jobs into three columns. Then compare with where a full-automation pitch would put them, and where we would.
| Job | AI does it | AI drafts, human decides | Human only |
|---|---|---|---|
| Topic research and trend scanning | |||
| Turning a voice note into a first draft | |||
| Choosing what you stand for | |||
| Editing a draft into your voice | |||
| Repurposing one long piece into five short ones | |||
| Writing replies to comments | |||
| Scheduling and publishing | |||
| Approving anything that goes out under your name | |||
| Pitching podcasts and editors | |||
| Monthly report on searches and mentions | |||
| Direct messages to prospects | |||
| Checking facts, numbers and quotes |
The column people most often get wrong is the middle one. "AI drafts, human decides" is not a compromise; it is the design that gets most of the speed and almost none of the risk, as long as the human step is real and not a rubber stamp. Our guide on human in the loop design covers how to keep it real.
How fully automated personal branding fails
These are failure modes of the architecture, not of any particular product. Each one has a structural fix.
| Failure | What causes it | Structural fix |
|---|---|---|
| Invented details | The model fills gaps in its context with plausible facts | A source-of-truth file and an evaluation step that rejects new claims |
| Voice drift | Drafts are pulled toward the model's default style over time | A voice profile from real writing, re-tested monthly against fresh samples |
| Visible automation | Auto-replies and auto-comments that miss context | No automation on any message to a person, ever |
| Topic sprawl | A topic generator optimises for novelty, not position | Topics chosen by a human against one positioning line |
| Silent breakage | Scheduler or API errors nobody sees | Error handling and alerts, as in error handling that stops silent failures |
| Unmeasured value | No baseline, so no one can tell if it helped | A baseline before the build and a monthly report after |
The general version of this list is in how AI agents fail. Personal branding is a narrow case, but it has one property that raises the stakes: everything published carries a real person's name.
Why bother: the audience moved into the assistants
The reason to invest in this at all is where people now look people up. OpenAI reported ChatGPT passing 900 million weekly users in February 2026.
Company-reported. Weekly, not monthly, users.
Source: OpenAI, Scaling AI for everyone, 27 February 2026.
Being found inside those answers is a different problem from being found on Google. Assistants mention people and companies without always linking to them; Ahrefs measured how often a mention comes with a link.
Company brands, 31,000+ mentions. When there is no link, the name and its description are the whole payload.
Source: Ahrefs, AI citations vs impressions study (31,000+ brand mentions, six AI platforms), 26 November 2025.
So the system's output is not traffic; it is a correct, consistent description of a person across the sources assistants read. That is a data quality problem as much as a writing problem, and it is the part engineers are good at.
Person Engine Optimization as a data problem
We call the target Person Engine Optimization, PEO: making a person findable and correctly described by Google and AI assistants. Seen from the engineering side, PEO has the shape of a knowledge system in reverse. Instead of retrieving facts for a model, you are publishing facts so that other people's models retrieve them correctly.
One record
A single file of facts about the person: name, role, proof points, dates. Every output is generated from it or checked against it.
Many surfaces
Website, LinkedIn, author pages, speaker bios, podcasts. Each one is a rendering of the record, not a separate piece of writing.
Drift checks
A monthly comparison of what each surface says, and what assistants say, against the record. Differences become tasks.
This is also why we are cautious about letting a model write the record itself. The record is the ground truth. It is written and signed off by the person.
Reference architecture
- Source of truthA versioned file of facts and banned claims, owned by the person and one editor.
- CaptureTranscription of calls, talks and voice notes, tagged by topic. Personal data handled as in personal data in AI pipelines.
- GenerationPrompts versioned and tested like code; drafts for posts, bios, article outlines and show notes.
- EvaluationAutomatic checks for new claims, banned phrasing and length, then human review. See writing evals that catch real failures.
- PublishingApproved content scheduled to each surface; nothing posts without approval.
- MonitoringMonthly checks of each surface and of assistant answers, compared against the record, reported to a person.
What it costs to run, as arithmetic
We do not quote a model price as current fact, because they change. The structure of the cost is stable: model calls for drafts and checks, a scheduler, storage, and the editor's time. Here is the arithmetic with a placeholder price you should replace.
| Item | Assumption (replace with yours) | Result |
|---|---|---|
| Drafts | 40 drafts a month at about 6,000 tokens each including context | 240,000 tokens |
| Checks | Two evaluation passes per draft at about 3,000 tokens each | 240,000 tokens |
| Model cost | At a placeholder of $5 per million tokens, blended | About $2.40 |
| Editor time | 15 minutes per draft | 10 hours |
The model is the cheap part. Human review time dominates, which is exactly why the design puts people only where they change the outcome. For the method, see token cost arithmetic.
If you are deciding whether to fund this
Three guides cover the questions that usually come next: what AI automation actually costs to build, the three-year cost of owning it, and whether to build in house, with an agency or a freelancer. For one person's personal brand, the honest answer is often a small build and a part-time editor, not a platform.
Who recorded the videos
The videos are by Bhavik Sarkhedi, the personal branding strategist whose method these systems implement. He is a three-time entrepreneur and the author of 21 books, has written for The New York Times, and holds a verified Google Knowledge Panel. His habit of using a model for the first and heaviest part of a task, then finishing by hand, is the design rule this page is built on.
"ChatGPT will be really helpful in doing the first 70% of your task."
More of his videos are at youtube.com/@bhavik_sarkhedi.
Questions about AI personal branding systems
What is AI personal branding?
Using language models to do the volume work behind a person's public presence, such as research, drafting, repurposing and monitoring, while people own the facts, the positioning and every human interaction.
Why not automate all of it?
Because the failures land on a real person's name. Models invent plausible details, drift toward a generic voice, and cannot hold a conversation as you. Each of those is cheap for the system and expensive for the person.
Can AI help me show up in ChatGPT answers about my field?
Indirectly. AI tools can help produce and maintain consistent, well-sourced material about you, which is what assistants draw on. They cannot submit you to an assistant, and nothing can guarantee a mention.
Which parts should I automate first?
Capture and reporting. Transcribing what you already say, and summarising what happened last month, save time with almost no risk to your name.
Do you build on n8n, Make or custom code?
Whatever fits the volume and the team who will own it. Our guide on n8n, Make or custom code explains how we choose.
How do we start?
Email the team from this page with a short description of how your content gets made today. We reply with the split we would recommend and why.
Send us your split. We will tell you where it breaks.
Sort the twelve jobs above, or just describe your current setup in a few lines. We reply with the seams we would move and the checks we would add.
Email the teamOr write to wr.sarkhedi@gmail.com. The email opens with a short note already written; add a line about you and send.
Related engineering reading
- ChatGPTalker home
- The LinkedIn personal branding workflow
- All services
- Custom LLM applications
- RAG and knowledge systems
- Evaluation and guardrails
- Content production engine
- How AI agents fail
- Designing the human in the loop
- Evals that catch real failures
- Personal data in AI pipelines
- Error handling for automation
- Token cost arithmetic
- n8n, Make or custom code
- Keeping humans skilled