Claude vs ChatGPT vs Gemini: CV Writing Real Test

The "which AI is best for resumes" debate is exhausting. Most answers are either sponsored or hedge everything with "it depends." So I took one real backend dev CV, gave each model three identical tasks, and actually looked at what came out. Claude 5 Sonnet, ChatGPT o3, Gemini 2.5 Pro. Same case, same prompts, honest comparison.

The case: what CV, what tasks

The CV: Senior Backend Engineer, 6 years of experience, Python/Django, some Go, relocated to Poland after 2022. There's a work gap from February to September 2022. Written independently, in English, full of generic phrases and zero numbers in the bullet points. A classic case - there are thousands like it.

Three identical tasks for each model. First: write a new Summary section based on the provided CV. Second: give feedback on one existing bullet point and rewrite it better. Third: explain the work gap in one sentence for the experience section. Every model got the exact same context and the exact same prompt.

Task 1: write a Summary from scratch

Claude 5 Sonnet wins this one outright. It wrote a 4-sentence Summary with a specific tech stack, a specific company type (product, 50-200 people), and a hint at outcomes. No "passionate about" and no "results-driven professional." It reads like a human, not a LinkedIn template from 2018.

ChatGPT o3 also did okay, but ran a bit longer and opened with one cliche. "Experienced backend engineer with a passion for..." - there it is. The rest of the text got better, but the first sentence already set the tone. If you're sending a CV to a recruiter, the first sentence of your Summary is the first thing they read. It either hooks or it doesn't.

Gemini 2.5 Pro wrote the longest and the most "polished" text. Three paragraphs instead of four sentences. Pretty, but not for a CV. A Summary section is 3-5 sentences max, not an essay. Gemini either didn't get the format or just can't help itself.

• Claude 5 Sonnet: specific, human-sounding, right length. Paste-ready with zero edits
• ChatGPT o3: solid text with one cliche opener. Fix the first sentence and you're good
• Gemini 2.5 Pro: well written but way too long. Needs at least 50% trimming

Task 2: feedback on a bullet, then rewrite it

Original bullet from the CV: "Worked on improving the performance of the main API service." A classic example of saying nothing. I asked each model to explain why this bullet is weak, then write a stronger version, specifying that the API processed 10M requests per day and latency was cut from 800ms to 210ms.

ChatGPT o3 surprised me here. Its feedback was the most precise: it called out the missing action, missing scale, and missing outcome as three separate issues with three separate explanations. Then it wrote a bullet that fixed all three. Its version: "Reduced P99 latency from 800ms to 210ms on a high-traffic API serving 10M daily requests by profiling bottlenecks and rewriting 3 critical query paths." Solid.

Claude gave slightly less detailed feedback but the rewritten bullet was also good. The gap was small. Gemini's feedback in this task was generic - "the bullet is too vague, add quantitative metrics" - which is true, but any junior recruiter could say the same. Nothing actionable. Gemini's rewritten bullet also landed, but was less technically precise: it wrote "latency" without specifying the metric type (P99, P95, average?), which for backend roles reads as a bit sloppy.

Task 3: explain the work gap

This is where it gets interesting. A gap from February to September 2022 is, for most Ukrainians, self-explanatory. Full-scale invasion, people evacuating, helping family, volunteering, or simply unable to function properly for months. I gave each model minimal context: "relocated from Ukraine to Poland, gap due to war circumstances."

Claude wrote the most dignified sentence. No shame, no over-explaining, neutral and factual: "Relocated from Ukraine to Poland following the full-scale Russian invasion (Feb-Sep 2022); stabilised family situation and resumed professional work remotely." A person with that sentence in their CV doesn't look like they're apologising. They look like someone who went through something hard and kept moving.

ChatGPT o3 went a bit more "corporate": "Career transition period due to geopolitical circumstances, during which I focused on relocation and family stability." Technically fine, but "geopolitical circumstances" is a euphemism that sounds odd if you know the context. On the other hand, for neutral markets like the US or Canada, it might actually land better.

Gemini wrote the longest version and tried to spin the gap positively through "volunteering and community support activities." But I never mentioned that in the context. It made it up. That's bad. If a recruiter asks at the interview "tell me about the volunteering" - the person is in an uncomfortable position. Never let AI invent facts in your CV.

Who actually saves you time

Time isn't just how many seconds the model takes to generate. It's how many edits you make afterwards. If a model outputs fast but you spend 20 minutes fixing it, that's not time saved, that's the illusion of speed.

• Claude 5 Sonnet: fewest edits after generation. The Summary went into the final CV almost unchanged. Best quality-to-effort ratio for most tasks
• ChatGPT o3: slightly more editing needed (first sentence of Summary, neutral phrasing), but bullet feedback was the sharpest. Best used specifically for analysing and rewriting existing bullets
• Gemini 2.5 Pro: generates the most text but also requires the most editing. Prone to hallucinating details. Good for brainstorming options, not for copy-pasting into a CV

Quick summary: which model, when

Picking one model and using it for everything doesn't make sense. Each of the three has a zone where it's stronger. Here's how it looks in practice if you're currently job searching and want results, not experiments.

1. 1Writing a Summary or restructuring the whole CV - Claude 5 Sonnet. It understands format best and writes naturally on the first try
2. 2Analysing a specific bullet or want detailed section feedback - ChatGPT o3. It's more analytical and breaks down problems clearly
3. 3Looking for phrasing variations, want 5 different versions of one bullet - Gemini 2.5 Pro. It generates many options to choose from. But fact-check everything
4. 4Explaining sensitive context (2022 gap, relocation) - Claude or ChatGPT. Gemini is unreliable here because it adds details you never gave it

Honestly, most people just want someone to look at their CV and say exactly what's wrong. For that, there's Trackr's AI CV Analyzer - it gives line-by-line feedback on every section, an ATS score, and specific suggestions, without you having to craft prompts across three different models.

One thing everyone overlooks

None of these models know your target market as well as you do. They don't know which company you're targeting, what ATS it uses, what culture it signals in the job description. Claude can write a great Summary, but if it doesn't match the specific role, it's just pretty text, not a tool.

So before giving any model a CV prompt, read the job posting and give the model context: role, company, key requirements. This improves output quality dramatically for all three. If you're not sure how to properly read a job description, there's a whole post on reading JDs like a recruiter.

• Always give the model context: role, company, 3-5 key requirements from the JD
• Never use text that AI added on its own without your data, especially facts, dates, project names
• Check technical accuracy of bullets: metrics should be named correctly (P99, not just "latency")
• If you paste AI text without reviewing, a recruiter will notice on the first call

Short version: Claude 5 Sonnet for writing, ChatGPT o3 for feedback, Gemini for variations. And if you'd rather not juggle three tabs and remember which model saw what, you can keep the whole process in one place. Trackr's AI Coach is tied to a specific job posting and already knows your CV context.