How I Automated LinkedIn Without Losing My Voice

Case Study

Marketing is not, and never has been my passion or area of expertise. I'd even say it's the area of business I dread most.

 So, how was I going to start getting my name out there without burning out trying to churn out content for LinkedIn? How much of the process could I really automate with AI tools without losing my personality? 

 This portfolio project is very personal because it is an example of how I use AI in my own work, including:

  • the real pain point,

  • actual prompts, 

  • what I trust (and what I don't), and 

  • the results. 

Spoiler Alert

 The automated system worked:

30/36

posts complete with proofreading only

3 months

of planned content from one reusable workflow

~1 hour/ week

to draft, create graphics and post 3 times a week.

Where human review mattered:

  • missing personal context was correctly left incomplete;

  • one location was inferred incorrectly from my CV;

  • some text needed tone/clarity edits;

  • two sentences were unclear enough that I rewrote them manually;

  • the addition of humour to behind-the-scenes posts.

Why I Chose ChatGPT

For this project, I decided to use ChatGPT. 

 I designed the project with some specific technical criteria in mind:

  • Plan first - the project and process were carefully planned before I started any technical work. 

  • Trusted source material - ChatGPT enabled me to organise work into projects and build a collection of documents to use as my source material. This keeps the AI focused on my content (reducing the risk of invented facts or hype).

  • Auditable steps — the workflow was split into clear sequential stages, from source material and strategy through copy, design and the final production pack, with human review between each stage before publication.

  • Human review at every stage – I didn’t start the next stage until I had manually reviewed the output from the last and checked for accuracy, tone, content balance, etc.

  • Use Canva consistently - the early decision to use Canva for graphic creation allowed for more efficient image creation for posts using a brand style guide.

  • A sustainable workload - I set a goal of a sustainable 3-post-per-week frequency to protect energy and prevent burnout.

What I Gave It To Work With

  • Drafts of my website copy for landing page and existing portfolio projects. 

  • Brand style sheet (branding colours, typography, etc.).

  • Summary document of my career and education highlights.

  • My CV

  • A course brochure for the University of Cambridge course I completed on Data Science and Machine Learning.

How I Designed The Process

Source Material →

01

Strategy & Review →

02

Copy & Review →

03

Design & Review →

04

Production Pack →

05

Final Human Review →

06

Publish

07

Human review is built into the process rather than bolted on at the end.

The Prompts

01

Prompt 1: Plan The Content

  • “You are a marketing expert. I'm launching as a data science freelancer and need a realistic LinkedIn content plan for the next three months. Please use the source files, which include my website copy, portfolio projects, CV and a brochure for a recent course I completed, as inspiration, and identify gaps I can fill with LinkedIn posts.  Please include some posts that are human/ context focused (networking invites, who am I, where am I based, what I do in spare time, single mum, why freelancing, etc.)”

Review and edit

  • Checked for a balanced mix of technical authority and human/context-focused posts.

  • Verified that the 3-post-per-week schedule has a good balance of content and themes.

  • Added missing context (I hadn’t included anything about my personal motivators or lifestyle in the source material)

02

Prompt 2: Draft The Posts

  • “You are a copywriter specialising in professional and approachable LinkedIn content.

    For each post in my content plan, use the source files to draft a strong hook, the main content, and a call to action that match my content, personality and voice.

    Do not invent facts, opinions, experiences, results or personal details. If a post depends on something that has not happened yet, leave a clear placeholder rather than making it up.

    I want the content to build on the principle of know, like and trust, showing strong communication and commercial judgement alongside technical expertise.

    Optimise each post for LinkedIn reach where this can be done naturally, without compromising value, readability, authenticity or uniqueness. In particular:

    • make the topic or business problem clear early;

    • use relevant terms naturally rather than keyword stuffing;

    • favour specific, recognisable problems over generic advice;

    • use hooks that create genuine interest without clickbait;

    • make posts easy to read and worth finishing;

    • use calls to action that encourage relevant, low-effort conversation rather than engagement for its own sake.

    Do not force every post into the same structure or length. Human and personal posts should still feel human, and not every post needs to teach a business lesson.

    Avoid jargon, hype, buzzwords, generic LinkedIn phrasing, exaggerated claims and formulaic engagement bait.”

Review and edit

  • Ensured tone matches website voice: clear, logical, grounded, and jargon-free.

  • Verified adherence to the 'know, like, and trust' framework without inventing metrics or opinions.

  • Filling gaps: ChatGPT left some text incomplete where it did not have enough context or details. The prompt and workflow design successfully helped prevent the invention of information.

  • Some additional information was needed (for example, ChatGPT had inferred my location based on CV content, and networking posts were hypothetical, whereas I preferred an invitation to meet).

03

Prompt 3: Build the Design Templates

  • “You are a design expert creating a practical, repeatable LinkedIn visual system for a solo data science freelancer.

    Use my LinkedIn content plan, copy drafts, source files and style sheet as the basis for the design system.

    The goal is to create the smallest reusable Canva template system that can support the full three-month content plan without making the feed look repetitive or over-designed.

    Prioritise:

    • low design workload;

    • visual consistency;

    • readability on mobile;

    • professional but approachable presentation;

    • enough variation to support the know, like and trust content mix.

    Create a maximum of four reusable Canva master templates. These should cover the main visual jobs in the content plan, for example:

    • single-idea statement graphic;

    • educational/framework carousel;

    • case-study/results carousel;

    • results/reflection or process carousel, only if genuinely needed.

     

    For each master template, define:

    1. its purpose;

    2. which types of posts it should be used for;

    3. page/slide structure;

    4. fixed layout rules;

    5. typography hierarchy;

    6. colour use;

    7. image or icon treatment;

    8. CTA placement;

    9. maximum recommended text per slide;

    10. rules for keeping repeated use visually fresh without redesigning the template.

     

    Use my branding colours and fonts from the style sheet in sources.

     

    Visual direction:

    • editorial, intelligent, calm and approachable;

    • generous white space;

    • strong hierarchy;

    • restrained use of colour;

    • simple shapes, lines and diagrams where useful;

    • no generic AI imagery;

    • no glowing brains, robots, circuit boards or neon gradients;

    • no fake dashboards or decorative charts;

    • no unnecessary stock imagery;

    • no excessive icon use.

    The visual system should support my positioning as an experienced data specialist who communicates clearly and pragmatically, rather than looking like a generic technology brand.

     

    Generate the prompt for each design asset that can be used in Canva.”

Review and edit

  • Aligned visual concepts with brand style guide parameters for Canva templates.

  • Formatted slide-by-slide text overlays for PDF carousels.

  • Templates actually look and feel right

04

Prompt 4: Pulling Everything Together

  • “You are a LinkedIn content production editor.

    Use my approved LinkedIn content plan, approved post copy, approved Canva master templates, source files and style sheet.

    Your job is not to rewrite the strategy. Your job is to assemble a complete, practical production pack for each post so I have everything I need in one place to create and publish it.

    For each post, output the following fields in this exact order:

    1. Week / day

    2. Post title / topic

    3. Know / Like / Trust role

    4. Format type

      • TEXT

      • PHOTO

      • 01 EDITORIAL STATEMENT

      • 02 FRAMEWORK

      • 03 CASE STUDY

    5. Visual asset / image direction

      • For TEXT: write “None”

      • For PHOTO: describe the type of personal/local image to use

      • For Canva templates: identify the approved template and any relevant variant

    6. Canva/template content

      • For TEXT or PHOTO posts: write “Not required”

      • For 01 EDITORIAL STATEMENT: provide the label, headline, supporting line and any emphasis instruction

      • For 02 FRAMEWORK: provide slide-by-slide copy

      • For 03 CASE STUDY: provide slide-by-slide copy

      • Do not redesign the approved template

      • Do not repeat brand/layout specifications unless an exception is genuinely required

    7. LinkedIn hook

      • Use the approved hook unless a minor edit is needed to fit the final format

      • Do not make it more sensational or clickbait-driven

    8. LinkedIn post text

      • Use the approved copy

      • Preserve my tone: clear, professional, approachable, pragmatic and jargon-free

      • Do not invent facts, opinions, results or experiences

    9. Call to action

      • Use the approved CTA

      • Keep it relevant and low-friction

      • Do not force a question if the post works better without one

    10. Notes before publishing

    • Flag any placeholder, future metric, image requirement, fact-check, local detail or manual update needed before publication

    • If nothing is needed, write “Ready to publish after normal review”

    1. Source grounding

    • Briefly note which source file(s) support the factual content

    • If part of the post comes from personal context supplied directly by me rather than the files, state that clearly

    Important rules:

    • Keep all information for one post together.

    • Do not scatter copy and design instructions across separate sections.

    • Do not generate a new graphic concept if an approved template already fits.

    • Do not duplicate the full LinkedIn caption inside the Canva asset.

    • Do not invent project results or future reflections.

    • If a future post depends on evidence that does not yet exist, keep the approved placeholder and clearly flag it in “Notes before publishing”.

    • Preserve the distinction between human posts and technical posts: personal posts should not be over-designed or forced into business lessons.

    • Use British English.

    Output the full three-month plan grouped by week.

    Use this exact heading structure for every post:

    Week X — Day — Post title

    Know / Like / Trust:
    Format:
    Visual asset / image direction:
    Template:
    Canva/template content:
    Hook:
    Post text:
    CTA:
    Notes before publishing:
    Source grounding:”

Review and edit

  • Consistency of tone, messaging, feel

  • Placeholders correctly assigned where more information is needed.

  • Results are actually easy to read, understand and implement

Example Outputs

The first three posts.

Month 1 — Know me -Week 1

  • Know / Like / Trust: Know + Trust
    Format: PHOTO
    Visual asset / image direction: Current, approachable photograph of you. Natural light and a relaxed professional setting rather than a formal corporate headshot.
    Template: Not required
    Canva/template content: Not required

    Hook:
    I’ve spent more than a decade solving problems with messy, real-world data. Now I’m taking that experience freelance.

    Post text:
    My route into data science started with oceanography.

    From there, I moved into scientific data management, software development and operational data systems.

    That meant working with the sort of data that rarely arrives neat, complete or conveniently organised — and with systems that had to work reliably in the real world.

    More recently, I’ve expanded that foundation into modern data science, machine learning and AI.

    The technology has changed over the years.

    The part I enjoy hasn’t:

    understanding complicated information, finding out what is really happening, and turning it into something useful.

    I’m now building a freelance practice helping businesses untangle messy datasets, improve data-heavy processes and answer questions they haven’t been able to answer yet.

    You bring the knowledge of your business.

    I’ll handle the technical side — without unnecessary jargon.

    CTA:
    If we haven’t met before, say hello and tell me what you do.

    Notes before publishing:
    Choose a current photograph that feels recognisably like you rather than generic “consultant” imagery. Ready to publish after normal review.

    Source grounding:
    Professional experience and career progression are grounded in the CV and website copy.

  • Know / Like / Trust: Trust
    Format: 01 EDITORIAL STATEMENT
    Visual asset / image direction: Use the Cream-background variant of the approved 01 Editorial Statement master.
    Template: 01 EDITORIAL STATEMENT

    Canva/template content:
    Label: REAL-WORLD DATA
    Headline: Missing fields. Conflicting reports. Manual workarounds.
    Supporting line: Messy data doesn’t automatically mean useless data.
    Emphasis: Use Academic Burgundy on “Conflicting reports” or “Manual workarounds”, not both.
    CTA on asset: None.

    Hook:
    Missing fields. Conflicting reports. Manual workarounds. A spreadsheet nobody completely trusts.

    That’s what I mean by messy data.

    Post text:
    Real-world business data is rarely perfect.

    It might be spread across several systems.

    Two reports might supposedly measure the same thing but give different answers.

    Some information is missing.

    Someone created a manual workaround years ago that everybody now depends on.

    Or there’s an old database that works perfectly well — as long as nobody touches it.

    That doesn’t automatically make the data useless.

    Good data analysis starts with understanding those imperfections.

    What matters?

    What can be fixed?

    What limitations need to be accepted?

    And what can the data genuinely tell us?

    I spent years working with complex scientific and operational datasets, so I don’t expect perfect data.

    I expect reality.

    CTA:
    Do you have a spreadsheet, report or dataset in your business that everyone uses but nobody completely trusts?

    Notes before publishing:
    Ready to publish after normal review.

    Source grounding:
    The “Data Realist” positioning and experience with messy operational data come from the website copy and CV.

  • Know / Like / Trust: Like
    Format: PHOTO
    Visual asset/image direction: Prefer one of your own mushroom photographs. A genuine weightlifting or kickboxing photo is an alternative if you already have one you like. Do not create a branded collage merely to show all four interests.
    Template: Not required
    Canva/template content: Not required

    Hook:
    Things you’re more likely to find me doing when I’m not staring at data:

    Post text:
    Spending time with my son.

    Weightlifting.

    Taking far too many photographs of wild mushrooms.

    Or attempting kickboxing.

    That last one is definitely still a work in progress.

    And yes, mushroom photography is every bit as niche as it sounds.

    No metaphor about mushrooms being like data.

    No lesson about kickboxing making me a better consultant.

    Sometimes hobbies can just be hobbies.

    CTA:
    What’s your slightly niche thing outside work?

    Notes before publishing:
    Use your own photograph. No text overlay or branded frame. Ready to publish after normal review.

    Source grounding:
    Your hobbies and spending time with your son are stated in the website copy.

So, Did It Work?   

What It Produced

  • 30 of the 36 posts generated were complete and only needed proofreading; 6 posts needed more input (often because they were updates on events or work that hadn’t happened yet).

  • The output included real diversity and depth of content and felt very aligned to my messaging and voice. I was pleasantly surprised by how authentic it felt.

What It Saved Me

  • The process saved hours and preserved creative energy during a high-stress launch phase (see below for a realistic breakdown of time spent).

  • The process very efficiently repurposed existing work (website copy and portfolio projects) into content for a new channel – this is now a reusable, repeatable method for future content. 

What I Learnt About The Process

  • Detailed prompts and breaking the process into phases with manual quality control at each stage meant transparency and confidence in the process. 

  • Three of the four prompts needed minor edits to get output that met my expectations. ChatGPT itself recommended several of those edits.

Where AI Got It Wrong

The review process identified:

  • Information that wasn’t available (the more personal posts that weren’t included in my website copy or CV),

  • That my location had been inferred from my CV and was factually incorrect.

  • Occasional editing to change tone or make the message clearer.

  • Two instances of sentences where I didn’t understand what the message was and so manually rewrote.

  • The addition of humour to behind-the-scenes posts.

The Unexpected Result

  • This project fundamentally shifted LinkedIn from a dreaded chore into a motivating and inspiring systematic process.

A Realistic Timeline

I already had all of my source material ready, so in reality, this ongoing project looks like:

One-off – 3 hours                       

  • Design the process

  • Create the templates in Canva

Every three months – 5 hours

  • Review success of previous posts

  • Adapt and run prompts, review and edit

Once a week – 1 hour

  • Create 3 graphics using templates and schedule on LinkedIn with text (hook, text, and call to action)

  • I'm using my own copy, words, and tone as my source material, not trying to reproduce or copy someone else's work. 

  • I'm being open about my use of AI (including writing this portfolio project published on my website). 

  • Manual review at every stage of the process to prevent hallucinations or misunderstandings from diminishing the accuracy of my posts. 

The Ethical Rules I Set For Myself

What I’d Automate - and What I Wouldn’t

AI tools are powerful, but my philosophy is that they are most useful and effective when the process is carefully designed, supervised and curated by someone who understands the strengths and limitations, in this case, by:

  • designing the workflow before choosing the technology,

  • providing trusted source material,

  • separating tasks into stages,

  • reviewing outputs from each stage manually,

  • using AI to reduce effort rather than surrender judgement.

How Can I Help?

I design processes around real business problems, ground them in trusted information, build in sensible human checks, and make them repeatable.

If you have a repetitive, text-heavy or decision-heavy process that feels harder than it should, I can help you work out whether AI is actually useful — and, if it is, design a workflow that keeps you in control.

This might include:

  • turning repetitive admin into a structured AI-assisted process;

  • summarising or categorising large volumes of text;

  • repurposing existing business content across channels;

  • creating internal knowledge or research workflows;

  • helping decide where AI should — and should not — be used;

  • building human-in-the-loop processes where accuracy and judgement matter.


If you are curious about building practical workflows for your business or launch, let's have a conversation - no hard pitch, just a friendly chat about what is possible.

Let's have a chat! →

Intrigued by the types of services I offer? Find out more on my services page.

See some example services →