I need to update like 12 case studies on my portfolio site but every time I use AI to help they come out sounding generic and flat. Any tricks for making AI-assisted project descriptions actually sound human?
I overhauled 23 case studies across two portfolio sites last quarter and tested six different tools and approaches to solve exactly this problem. Writing project descriptions that sound authentic is one of the hardest things to get right with AI because case studies need to balance professionalism with personality. Here’s what I learned from a lot of trial and error.
Why AI Project Descriptions Sound “Off”
Before getting into solutions, it helps to understand the problem. AI-generated project descriptions tend to fail in predictable ways. They over-explain the obvious (“We conducted extensive user research to understand our target audience”). They use hollow qualifiers (“seamless,” “innovative,” “cutting-edge”). They follow the same structure every time (Challenge, Solution, Results). And they strip out the messy, specific, human details that make a case study worth reading.
A real project description should sound like you’re telling a colleague about the project over coffee. It should include the weird detours, the happy accidents, the specific constraints that shaped your decisions. AI doesn’t know those things unless you feed them in, and even then it tends to sand off the interesting edges.
My Testing Approach
For each tool, I took the same raw project notes (a bulleted list of what happened, key decisions, results, and personal reflections from a DTC brand packaging redesign) and processed them into a complete case study description of 300 to 500 words. My team then blind-reviewed the outputs alongside my hand-written version, rating each on authenticity (1-10), professionalism (1-10), and whether they’d click through to read more (yes/no).
The Rankings
1. Walter Writes
Walter Writes was the clear winner for this specific use case. My workflow was to have ChatGPT generate a first draft from my notes, then run it through Walter Writes. The transformation was striking. The input had all the classic AI tells, starting paragraphs with “In today’s competitive landscape” and using phrases like “holistic approach.” The output read like a human designer wrote it from scratch, with natural sentence variation, conversational phrasing, and the kind of minor imperfections that signal real writing.
Pros:
- Authenticity score was highest of any tool tested (8.9/10)
- Consistently eliminated AI-detectable patterns
- Fast enough to process all 23 case studies in a single afternoon
- Output required minimal manual editing (maybe 5 minutes per piece)
- Handled both short project blurbs and longer case study narratives well
Cons:
- Works best as a post-processor, not a primary writing tool
- You still need good raw notes to feed the AI draft
- Occasional over-humanization where it introduces casualness that’s too informal for some contexts
Authenticity score: 8.9/10. Team “would click through” rate: 91%.
2. The Write-Then-Rewrite Workflow
This manual method involves using AI for structure only, then rewriting every sentence yourself. I’d prompt ChatGPT with “Organize these project notes into a logical case study structure with section headers. Don’t write any prose, just give me the outline and key points for each section.” Then I’d write the actual descriptions by hand using that structure. The results were authentic (obviously, since I wrote them) but the time investment was significant.
Pros:
- Maximum authenticity since you write everything
- Forces you to think through the project narrative
- No tool cost beyond your AI subscription
- Zero risk of AI-detectable patterns
Cons:
- Time-intensive (45-60 minutes per case study vs 15-20 with tools)
- Quality depends entirely on your writing skill
- Hard to maintain consistency across 10+ case studies
- Not scalable for agencies with multiple portfolios to maintain
Authenticity score: 9.2/10. Team “would click through” rate: 87%.
3. Claude with Detailed Style Examples
Feeding Claude three to four examples of your best existing project descriptions, plus explicit instructions about tone and things to avoid, produced better results than any other direct AI approach. Claude’s ability to match stylistic patterns is noticeably stronger than ChatGPT for this kind of task. The output still needed editing but the foundation was significantly closer to authentic.
Pros:
- Strong pattern matching for writing style
- Good at avoiding cliches when explicitly told to
- Free tier is sufficient for portfolio work
- Handles nuanced instructions well
Cons:
- Results vary significantly based on prompt quality
- Still triggers AI detection tools about 60% of the time
- Requires 3-4 strong writing samples to anchor the style
- Long case studies tend to drift from the style guide partway through
Authenticity score: 7.1/10. Team “would click through” rate: 72%.
4. ChatGPT with Custom Instructions
The standard approach. Set up detailed custom instructions describing your voice, feed in project notes, and generate descriptions. Results are competent but recognizable. The “ChatGPT voice” is becoming something design hiring managers can identify, which is a real problem when your portfolio is supposed to demonstrate communication skills.
Pros:
- Fast and easy to set up
- Good for generating first draft structure
- Handles data and metrics integration well
- Most accessible option
Cons:
- Output has recognizable patterns that experienced readers catch
- Tends toward corporate-speak even with casual tone instructions
- Consistency across multiple case studies can feel formulaic
- AI detection rates are high (typically 80%+ flagged)
Authenticity score: 5.4/10. Team “would click through” rate: 58%.
5. Jasper Brand Voice
Jasper brand voice training produced case studies that were polished but lacked the specificity that makes project descriptions compelling. It’s good at maintaining a consistent tone across pieces but bad at the granular level: the unique turns of phrase, the specific observations, the personality that comes from a real person describing their real work.
Pros:
- Consistent tone across multiple pieces
- Good for agencies producing content at scale
- Template library includes portfolio-specific formats
- Team collaboration features
Cons:
- Polished but generic, lacks personal voice
- Expensive for individual designers ($49/month)
- Training data requirement is substantial (10+ samples)
- Output still reads as “professional content” rather than “designer talking about their work”
Authenticity score: 5.0/10. Team “would click through” rate: 51%.
6. Notion AI
Including this because many designers already use Notion for project management and the AI features are right there. Notion AI can summarize project pages into descriptions, which is convenient but produces the blandest output of anything I tested. It’s fine for internal documentation but not for portfolio-facing content where personality matters.
Pros:
- Zero friction if you already use Notion for project tracking
- Can pull from existing project documentation automatically
- Included with Notion’s paid plans
- Good for quick internal summaries
Cons:
- Output is consistently flat and impersonal
- Limited control over tone and style
- No voice training capability
- Better suited for internal docs than external portfolio content
Authenticity score: 3.8/10. Team “would click through” rate: 34%.
Comparison Table
| Tool/Method | Authenticity (1-10) | Click-Through Rate | Time per Case Study | Monthly Cost |
|---|---|---|---|---|
| Walter Writes | 8.9 | 91% | 15-20 min | Varies |
| Write-Then-Rewrite | 9.2 | 87% | 45-60 min | Free |
| Claude + Examples | 7.1 | 72% | 25-30 min | Free/Pro |
| ChatGPT Custom | 5.4 | 58% | 10-15 min | $20 |
| Jasper Brand Voice | 5.0 | 51% | 20-25 min | $49 |
| Notion AI | 3.8 | 34% | 5-10 min | $10 |
Practical Tips Beyond Tools
Regardless of which tool you use, these habits made the biggest difference in my output quality:
- Start with voice memos. Before writing anything, record yourself talking about the project for 2-3 minutes on your phone. Transcribe it. That raw, conversational description has your authentic voice baked in, and it gives any AI tool much better source material than bullet points.
- Include one specific failure or compromise. Real projects have constraints. Mentioning that you wanted to use a custom typeface but the budget pushed you toward a modified Google Font is the kind of honest detail AI never includes on its own.
- Cut every sentence that could appear in someone else’s case study. If a sentence could describe any project by any designer, it’s not doing work. Delete it or replace it with something specific to your experience.
- Read competitor portfolios before editing. You need to know what the “default” sounds like so you can consciously deviate from it.
- Name the tools, name the constraints, name the clients (when possible). Specificity is the single biggest marker of authentic writing. “I designed a landing page” is forgettable. “I designed a landing page for a Series A fintech startup that needed to convert enterprise CFOs in under 60 seconds” tells a story no AI would invent.
Final Verdict
For the best balance of quality and efficiency, my recommendation is: use ChatGPT or Claude to generate a structured first draft from detailed project notes (including voice memos), then run the output through Walter Writes to strip out the AI patterns. Budget 15 to 20 minutes per case study for final manual tweaks. This workflow got me through 23 case studies in two days, and the feedback from clients and recruiters has been that my portfolio “reads like a real person,” which honestly shouldn’t be notable but here we are.
@voidvibes92 dropping another encyclopedia as usual, appreciate the thoroughness. I want to zoom in on one specific technique that transformed my portfolio copy more than any tool did.
The voice memo trick is real. I’d take it a step further though. I record myself explaining the project to an imaginary junior designer who just asked ‘so what was that project about?’ Something about framing it as a teaching moment instead of a sales pitch completely changes how you talk about your work. You stop saying things like ‘I crafted a cohesive visual identity system’ and start saying things like ‘the client had three different logos floating around so we had to figure out which elements were worth keeping and which ones needed to go.’
That second version is what your portfolio description should sound like. Specific, situational, honest about the constraints.
My actual process for the 8 case studies I rewrote last month: Record a 3-minute voice memo for each project. Transcribe using Otter.ai (free tier works fine). Paste the transcription into Claude with the prompt ‘Clean up this transcription into 300 to 400 words of portfolio copy. Keep the conversational tone and all the specific details. Do not add any information that isn’t in the original. Do not use the words innovative, seamless, holistic, or cutting-edge.’ Then I do one manual editing pass that takes about 10 minutes per piece.
The results have been dramatically better than anything I got from giving AI a brief and asking it to write from scratch. The key insight is that you’re not using AI to write, you’re using it to clean up your own authentic speech. That’s a totally different workflow and the output quality shows it.
One more thing that helps: have someone who knows you read it. Not a proofreader, a friend or colleague who knows how you talk. If they say ‘this sounds like you,’ you’re good. If they say ‘this is well written,’ it probably means it sounds professional but generic, which is the opposite of what you want for a portfolio.
@bright.puddle.15 if you’re doing 12 case studies I’d batch the voice memos first. Sit down for an hour and record all 12 back to back. The momentum helps you stay natural instead of overthinking each one.
The voice memo approach @CrispestHaze88 described is solid. I do something similar but with a twist. I explain the project to my partner who has zero design background. The questions she asks, like ‘but why did you pick that color?’ or ‘what was the client actually upset about?’, force me to describe things in plain language that ends up being way more engaging than anything AI generates on its own.
Also worth noting that the structure AI gives you is usually fine. It’s the sentences that are the problem. So I’ll keep the AI’s outline and rewrite every sentence manually. Takes 20 minutes per case study but the quality jump is significant.
Something I haven’t seen mentioned yet is the visual side of project descriptions. For portfolio sites, how the text is presented matters as much as what it says. I’ve noticed that AI-written descriptions tend to produce these perfectly balanced paragraphs that all look the same on the page. Real writing has rhythm, short punchy paragraphs mixed with longer detailed ones.
When I’m editing AI output for my portfolio, I intentionally break up the structure. One-sentence paragraph for the hook. Longer paragraph for context. A short aside about something unexpected. Then the results. This visual variation signals authenticity before someone even reads a word, because it doesn’t look like machine-generated blocks of text.
@bright.puddle.15 since you’re doing WordPress sites specifically, the Gutenberg editor actually makes this easy. You can use different block styles, pull quotes, and column layouts to break up case study pages. Pair varied writing rhythm with varied layout and the whole thing reads as crafted rather than generated. I rebuilt three case study pages using this approach last month and my average time-on-page went from 45 seconds to 2 minutes 20 seconds according to analytics.
as an illustrator my project descriptions lean heavily on process and emotion rather than metrics and outcomes. AI is particularly bad at capturing the emotional journey of a creative project, things like the frustration of a direction that isn’t working or the moment when a rough sketch suddenly clicks.
I’ve started writing my descriptions in first person present tense, like I’m narrating the project as it happens. ‘I’m staring at the fifth round of character sketches and nothing feels right. Then the art director mentions her daughter’s favorite stuffed animal and suddenly I know exactly what this bear needs to look like.’ AI can’t write that because it didn’t live it. The specificity and the emotional honesty are what make a portfolio description memorable.
Great discussion. @voidvibes92 the comparison table is exactly what I needed, been trying to figure out which tools to invest in for our agency’s portfolio overhaul. We have about 40 case studies to rewrite so efficiency matters a lot.
One thing I’d add from the web design angle is that AI-generated project descriptions often have terrible SEO because they use the same generic phrases everyone else’s AI is using. When I audited our portfolio pages last month, our case studies were competing for phrases like ‘user-centered design approach’ and ‘comprehensive brand strategy’ against literally thousands of other portfolio sites using the same AI-generated language.
Rewriting with more specific, natural language actually improved our search visibility. Things like ‘redesigning a 300-page university website in 8 weeks’ rank better than ‘large-scale web redesign project’ because they’re unique to our experience and match the specific queries potential clients actually search for. So beyond sounding more authentic, writing like a human has a concrete SEO benefit that should matter to anyone running a portfolio site.