Artificial intelligence has meanwhile become part of the professional toolkit. Tools like ChatGPT and other AI software are marketed as assistants that can draft, analyse, and even create. Yet for all the hype, working with AI often reveals sharp edges, blind spots, and limitations that are important to know.
Over the past months I have integrated ChatGPT into several projects, ranging from IT consulting to creative work. The results have been mixed — sometimes impressive, sometimes frustrating. Below I share some of the struggles, not to dismiss the technology, but to highlight the reality behind the promises.
Accuracy vs. Illusion of Accuracy
One of the most persistent (and annoying) challenges has been factual reliability. ChatGPT produces answers with second-to-none confidence, but that confidence often masks uncertainty. On technical topics such as programming, WordPress integration, or database queries, it sometimes delivers flawless snippets — and sometimes code that simply won’t run.
The issue is not only correctness, but also detectability. Errors are often wrapped in perfect grammar and convincing explanations. Spotting mistakes requires the same level of diligence I would apply when reviewing a junior colleague’s work. In practice, this means AI is more of a brainstorming partner than a source of unquestioned truth.
Memory and Context Gaps
Despite the label of a “conversation,” ChatGPT does not truly remember. Within a single session it can keep track of a discussion, but once the session ends, context is lost. This has forced me to repeatedly restate requirements, re-upload references, or remind the system of earlier agreements.
For long-term projects this lack of continuity is a real bottleneck. Building complex deliverables — such as structured documentation, multi-step WordPress plugins, or serialized creative drafts — requires a shared memory. Without it, conversations circle back, consuming more time than they save.
Creativity Under Constraints
Another limitation becomes clear in creative projects. Whether drafting a blog post, designing a thumbnail, or developing copy for a campaign, ChatGPT can generate endless variations. Yet the creativity is bounded by patterns in its training data.
Recently I asked for very specific layouts or stylistic features — for example, bold 3D text in thumbnails or alignment rules for branding. The results were inconsistent. Some outputs captured the request perfectly; others ignored constraints or repeated mistakes I had already corrected. This inconsistency adds friction where precision matters most: corporate identity, design standards, and brand voice.
The Struggle With Nuance
Human communication is rich in tone and subtext. Asking ChatGPT to sound less “salesy,” more “journalistic,” or appropriately formal often requires multiple iterations. The model tends to over-correct, swinging from overly casual to stiffly corporate.
This becomes most visible in sensitive contexts such as press releases or philosophical essays. While the tool can generate grammatically correct English, the subtle calibration of tone — something a human writer can intuitively achieve — is still beyond its reach.
Time Saved vs. Time Spent
One of the paradoxes of working with AI is the trade-off between speed and reliability. Drafting an article or document with ChatGPT is undeniably faster than starting from a blank page. Yet the time saved on drafting is often spent on reviewing, editing, and correcting.
In some cases, the net result is positive — AI reduces the friction of starting. In others, especially when precision and factual rigor are critical, the editing overhead cancels out the initial gain.
The Human Factor
Perhaps the most important lesson is that ChatGPT is not a replacement for human expertise. It can assist, inspire, and accelerate, but it cannot replace judgment. My background in IT and consulting allows me to recognize when the model makes mistakes; someone without that foundation might unknowingly adopt flawed outputs.
The promise of AI as an “assistant” is real, but only when paired with strong human oversight. Left unchecked, its flaws can mislead more than they help.
My Summary
My experiences with ChatGPT have been both valuable and challenging. The tool shines in rapid idea generation, draft creation, and exploratory problem-solving. But it struggles with factual reliability, continuity, nuanced tone, and consistent application of constraints.
For professionals, the takeaway is clear: Treat ChatGPT as an assistant, not as an authority.
Its outputs should be starting points, not final products. With the right expectations, it can be a powerful addition to the toolbox. With the wrong expectations, it risks becoming a source of extra work — or worse, costly mistakes.
Artificial intelligence is not magic. It is software, trained on patterns, limited by design.
Knowing where those limits lie is the key to using it effectively (and sensibly).