My Experiments with AI for Creating Content (Text & Images)
This is a reflection on my experiments with AI in creating
content—text and related visuals.
These were done with the intent of understanding what
actually works.
My background to set the context to this article
I am a person of few words – drawing an analogy from the world of poetry,
I am a fan of Ghazals (short, precise couplets) vs Ballads (long-form
narratives).
Also, I prefer to work without too many constraints on me – tending
more towards the creative side rather than more regimented.
In view of this, with the advent of social media, people
sharing more were getting noticed. So, I had to express myself more – clearly
and consistently.
And, to put some of my content on marketplace platforms, I
had to put my content in far more regimented way than I am used to.
Hence, I needed help and before the advent of Artificial
Intelligence (AI), I was seeking help from content professionals.
Once ChatGPT and Gemini were launched, I began experimenting
with these.
Having dabbled with Expert Systems way back in late 1980s, I
knew that AI in 2020s should be capable of doing much more.
Over time, through multiple projects, iterations, and
reflections, that curiosity evolved into a working discipline—one that reshaped
not just how I create content, but how I think about it.
Sharing below are some of my experiences and learnings.
When AI Stopped Being an Outsourced Agent
My early interactions with AI were simple: ask, generate,
refine.
I used all the “prompt engineering” knowledge that was being
widely shared.
The outputs were impressive—but often generic. Structured
well, written cleanly, but not quite mine— in terms of writing style, uniqueness
of approach and tonality.
The real shift happened when I stopped asking AI to write
for me and started engaging with it differently.
I started drafting my versions and asking AI to either
shorten or elaborate or make it more impactful.
Then, I would work on one section at a time e.g. title,
opening, body and closing.
I would see what AI has generated and then revise it to my liking
and resubmit for polishing or sanity checking.
This started working well and AI moved from being an
“outsourced agent” to a “thinking buddy” for me.
Earlier, I was reacting to AI. Now, I was directing it.
When AI Gets the Fundamentals Wrong
Invariably, AI would typically make assumptions that a
domain expert would make, and share the content or strategy based on this.
For example, in one of the early content explorations, AI
naturally structured a series of posts for social media as if the audience
would consume them sequentially.
That is a standard assumption. And a flawed one.
On social media platforms, not all posts appear on
everyone’s feed. Hence, people rarely see content in sequence. Yet, AI
defaulted to that pattern because it is statistically common.
That moment was important—not because AI was wrong, but
because I could recognise its pattern.
And once I picked it up, I could challenge it.
That led to a clear shift:
- Every
post must stand on its own
- Context
cannot depend on previous posts
- Curiosity
must be self-contained
A small structural change—but a fundamental strategic shift.
However, a point to be noted was that AI was quick to
acknowledge the flaw and work with me on correcting it.
There were other instances where even simple taxation principles
were applied incorrectly and I had to flag these up.
Hence, pick up patterns or other fundamental flaws and challenge AI. Don’t become its slave. Use it like a human buddy – one that augments your thinking but can make mistakes.
Precision Over Polish
AI is naturally inclined toward polished language.
But in my work, polish is not always the goal—precision
is.
There were multiple instances where I chose phrasing that
was:
- Slightly
unconventional
- Not
perfectly “correct”
- But
far more aligned with intent
For example, choosing to say “each one is unique” instead of
a more formally complete sentence—because the brevity creates sharper impact
and better visual balance, even if it slightly bends grammatical expectations.
This is where human judgment becomes critical.
Good content is not always grammatically optimal.
It is contextually right.
Visual Work: Challenges with Focus and Execution
When working on visuals—logos, icons, diagrams, conceptual
elements—the learning shifted.
Initially, the challenge was not generating visuals—but
getting AI to focus on the right concept.
The early outputs revealed three consistent issues:
·
The focus was on the body of the content, not
the core idea or title
·
The visuals were too loud, despite being
prompted to be minimalist and business-like
·
Iterations often drifted due to incorrect
interpretation of instructions
Once, this went on for 3 to 4 times. When out of
frustration, I mentioned that to AI in a prompt and asked if I was giving an
incorrect prompt. Also, asked it for suggestions to improve my input.
To my pleasant surprise, AI honestly acknowledged the
limitations of its model and accepted that my input was not the issue.
This led me to change my approach—from relying on AI to
generate finished visuals to using it to refine direction.
I started by using AI to explore possible themes. Then, I
created a rough draft myself and used AI to critique and improve it.
That shift—from generation to guided refinement—worked
significantly better.
Advantages of working with AI over humans
Amongst the disappointments and the struggles there were
some great positives of working with AI.
About couple of years earlier, I had worked with a marketing
consultant, a copywriter, and a graphics designer for evolving similar content.
That experience became an unintended benchmark and
comparisons are stark.
Quality and Quantity
With AI, both improved significantly within the same time frame.
The Fatigue Factor
With human collaborators, the process was different:
· They
would take a brief
· Do
their own interpretation and research
· Come
back with something that often felt clichéd or misaligned
Iterations would follow. And
then, inevitably:
· “Too
many changes”
· “We
need to close this now”
· “Further
revisions will cost more”
With AI, that constraint
disappeared.
There were no limits on
iterations. No negotiation on revisions. No friction in going back and forth.
The only limit was my own
patience to get it right 😀
The Ego Factor
One unexpected shift stood out.
Working with AI removed a layer
that is often invisible—but very real—when working with people: ego.
There was a certain mutual
openness in the interaction.
When outputs were repeatedly
off-track, simply calling it out led to:
· Acknowledgment
of limitations
· Transparency
about why it was happening
· Suggestions
for alternative approaches
There was:
· No
defensiveness
· No
negotiation over iterations
· No
friction from misaligned expectations
· No
“I am the expert” claims
That changed the nature of the
process itself. It became less about managing people and more about refining
thinking.
And that changed not just the
output—but the experience of creating it.
In hindsight, I was paying for their inability to fully understand my context.
The Discipline That Emerged
Over time, this stopped being a set of experiments and
started becoming a discipline.
A few principles shaped this:
- Engage, don’t outsource thinking
AI is your collaborator, not a substitute
- Question default patterns—especially AI’s
What is common is not always correct
- Design for how content is actually consumed
Not how AI assumes it is
- Use consistency selectively
Grammatical consistency matters—
but context of the content matters more.
Take ‘poetic liberties’ where needed.
- Shift between macro and micro levels of content
Start with the full picture, refine in parts, then review as a whole
- Iterate patiently
AI removes the cost of iteration—not the need for it
Over time, these stopped being principles —and became a way of working.
Final Thought
Artificial Intelligence does not define the quality of your
content.
Your Natural Intelligence does.
