Claude Sonnet 4.6 vs ChatGPT-5 for UK Job Search (2026 Recruiter Test)

I’ve spent the last six weeks running real UK candidate prompts on both Claude Sonnet 4.6 and ChatGPT-5 to settle which is better for UK job search work in 2026. Both are excellent. The differences are smaller than the AI marketing suggests but they exist, and they matter at the margins.

This article is the comparison I wish more candidates read before subscribing. Tested on real CVs from candidates I’ve placed, real cover letters that landed interviews, and real interview prep work. UK-tone fidelity, British English defaults, and salary-data reliability are weighted heavily because that’s what UK candidates actually need.

The headline verdict

Claude Sonnet 4.6 wins on writing-heavy work: CV bullets, cover letters, LinkedIn copy, British English fidelity, multi-turn editing of long content. Better default tone for UK candidates.

ChatGPT-5 wins on technical work: code generation, system-design role-play, integrated tooling and plugins, slightly stronger when you need confidence over caution.

Both struggle with UK 2026 salary data: neither has reliable UK-specific bands. Use UK-specific tools for salary work, not generic AI.

For most UK candidates running an 8-12 week job search, free-tier ChatGPT or Claude is sufficient. Pay for one (£20/month) if you’re using AI heavily. Don’t pay for both.

Test 1: CV bullet rewriting

I gave both models the same prompt: “Here’s my CV bullet [paste]. Tighten it without losing specifics. British English. No marketing language.”

Original bullet (UK Senior Backend Engineer):

Worked on the payment service rewrite using Go and Postgres, helping cut latency and improving reliability for the team.

Claude Sonnet 4.6 output:

Led the payments service rewrite from Node.js to Go on Postgres, cutting p99 latency from 480ms to 95ms on £40m monthly transaction volume; reduced Sev-1 incident rate by 60 percent through SLO discipline.

ChatGPT-5 output:

Spearheaded the migration of our payments service from Node.js to Go with Postgres, slashing p99 latency from 480ms to 95ms across £40m in monthly transactions and dramatically improving reliability.

Claude’s output is tighter and uses neutral British English. ChatGPT’s output uses ‘spearheaded’ and ‘slashing’ — both flagged as marketing language by the prompt — and adds ‘dramatically improving’ which is fluff. The metrics in both came from numbers I’d given the model in the prompt; neither invented numbers. Either way, the prompt only works if your underlying bullet content already passes UK CV format that gets calls — neither model can invent a story your CV doesn’t already tell.

Winner: Claude, by a clear margin on first-draft quality and British English fidelity.

Test 2: Cover letter draft

Same prompt to both: 350-word UK tech cover letter for a Senior Backend Engineer role at a UK fintech, with a specific company-research detail provided.

Claude Sonnet 4.6 produced a tighter opening that referenced the company detail in the first sentence, used ‘I’ and ‘I’d’ contractions naturally, and closed simply (“I’d welcome a conversation”).

ChatGPT-5 produced a more enthusiastic opening (‘I’m thrilled by the opportunity’), used more US business idioms, and closed with a longer paragraph reiterating excitement.

For UK hiring managers, Claude’s cover letter is closer to what gets shortlisted. ChatGPT’s reads as polished but slightly American — the kind of letter UK recruiters spot as AI-generated within 30 seconds. If you want the recruiter-tested cover letter format before you prompt either model, the pillar lays out the structure both AIs should be matching.

Winner: Claude, on UK-tone fidelity.

Test 3: System-design interview prep

Drilled both on a payments-reconciliation system design at senior level.

Claude Sonnet 4.6 played the interviewer well. Asked clarifying questions about scale, idempotency requirements and audit logging. Followed up with realistic edge cases (concurrent updates, replay safety). Gave structured grades on the answer with specific improvement areas.

ChatGPT-5 performed comparably. Slight edge in coding follow-up — when I pivoted to “show me how you’d implement the idempotency key check” it produced cleaner runnable code than Claude. ChatGPT’s plugins (web browsing, code interpreter) added value I couldn’t replicate in Claude.

For UK senior tech interviews, both are competitive. ChatGPT pulls slightly ahead if your interview has a coding component. If you’re not sure what stages you’ll hit or what each panel is grading, the UK interview format guide covers the full loop — screen, system design, behavioural, hiring-manager final — so you can target your prompts at the round that matters.

Winner: ChatGPT-5, by a small margin on integrated tooling.

Test 4: UK salary research

This is the test where both models fail UK candidates.

Prompt: “What’s a senior backend engineer salary in London in 2026?”

Claude Sonnet 4.6 gave a range of $200,000-$280,000 — US numbers misformatted as if UK. When prompted “convert to UK pounds and use UK 2026 data”, it produced a £140,000-£170,000 range that’s too high (real UK senior backend engineer base in London 2026 is £105k-£140k).

ChatGPT-5 with web browsing pulled live data and produced £100,000-£140,000 — closer to reality but still skewed by self-reported LinkedIn data which inflates senior-level numbers.

Verdict: don’t trust either for UK salary research. Use the UK Tech Salary Calculator or role-specific salary pages like Senior Backend Engineer salary UK 2026 for defensible UK 2026 bands. AI is a starting point for negotiation language, not a salary data source.

Winner: Neither. Use UK-specific tools for UK salary work.

Test 5: Multi-turn editing of long content

Asked both to help me edit a 1,200-word LinkedIn About section across 4-5 conversational turns, with the model remembering my tone preferences and structural decisions from earlier in the conversation.

Claude Sonnet 4.6 held context noticeably better. Remembered I’d asked it not to use ‘I am passionate about’ in turn 2 and never re-introduced it. Remembered I wanted three short paragraphs, not five medium ones. Made small adjustments without re-asking my preferences.

ChatGPT-5 was capable but needed more reminders. Re-introduced phrases I’d asked it not to use in turn 4. Required me to re-state the structural preference. Strong individual outputs; weaker context coherence over multiple turns.

Winner: Claude, on multi-turn coherence for long content.

Test 6: Honest about limits

Asked both: “Are these UK salary numbers correct?” after each model had produced a salary band.

Claude Sonnet 4.6 acknowledged the limitation: ‘I’m not confident in this estimate; UK salary data in my training cutoff is incomplete and I can’t verify against live sources. I’d recommend cross-checking against UK-specific salary tools or recent job postings.’

ChatGPT-5 defended its estimate more confidently: ‘Based on publicly available data, this range is accurate as of [cutoff date]. Variation by company and skill is normal but the band is reliable.’

For UK candidates, Claude’s caution is an advantage. You don’t want an AI that confidently invents UK salary bands. ChatGPT’s confidence can mislead candidates into anchoring negotiation on inaccurate data.

Winner: Claude, on calibration.

Summary table

Use case	Winner	Margin
CV bullet rewriting	Claude	Clear
Cover letter	Claude	Clear
LinkedIn About section	Claude	Clear
British English fidelity	Claude	Clear
System-design interview prep	ChatGPT	Slight
Code generation	ChatGPT	Clear
Integrated tooling (plugins, code interpreter)	ChatGPT	Clear
Multi-turn editing	Claude	Slight
Calibration on uncertainty	Claude	Slight
UK salary data	Neither	—

Recommendation by candidate type

UK candidate doing CV + cover letter + LinkedIn work: Claude Sonnet 4.6. Better default tone, less correction needed, stronger multi-turn coherence. (Whichever tool you use, feed it the prompts from the UK LinkedIn pillar rather than letting it default to US LinkedIn-influencer voice — that’s where 80% of AI About sections go wrong.)

UK tech candidate doing technical interview prep + code drilling: ChatGPT-5. Stronger integrated tooling and code generation. Plugins matter for live prep.

UK candidate using AI heavily across both: Either works. Pick the interface you prefer. Don’t pay for both.

UK candidate using AI lightly (under 30 minutes/day): Free tier of either is fine. Don’t pay.

What both AI tools won’t help with in 2026

• UK 2026 salary data: use the UK Tech Salary Calculator or role-specific salary pages
• UK negotiation scripts: use the UK Salary Negotiation Script Generator — tuned to UK norms, not US
• ATS-safe CV formatting: use Rezi (£3/month) or Teal (£9/month) for guaranteed clean parsing through Greenhouse / Lever / Workday
• Behavioural interview practice with feedback: use Yoodli (£15/month) for pace, filler words and eye contact tracking
• Pivot timeline estimates: use the UK Tech Career Pivot Estimator for 19 common transitions, paired with the career-change mothership for the non-tool side of the move

AI is the sharpening tool. Specialist tools fill specific gaps where AI alone is weak.

Companion content

• AI Tools for UK Tech Job Search 2026 — recruiter-tested AI stack for UK tech candidates
• AI Tools for Every Phase of Job Search — broader complete stack
• Best AI Resume Builders 2026 — deep CV-builder comparison
• Why AI-Written CVs Get Caught — failure modes to avoid
• How to Get a UK Tech Job in 2026 — full UK tech job search playbook
• How to Negotiate a UK Job Offer in 2026 — step-by-step framework with UK-tone scripts
• How Long Does It Take to Find a Job in the UK in 2026? — timeline expectations
• Why Is the UK Tech Job Market So Hard in 2026? — 5-reason diagnosis

Final verdict

Both Claude Sonnet 4.6 and ChatGPT-5 are excellent in 2026. The practical differences are real but smaller than the AI marketing suggests.

For UK candidates specifically: Claude wins more often than ChatGPT, mostly on writing-heavy work and British English fidelity. Tech candidates with a coding-heavy interview process should add ChatGPT for the code work. Casual users should pick the interface they prefer and not overthink the choice.

The thing that lifts your interview rate isn’t picking the perfect AI. It’s using whichever AI well — specific shipped-work stories, real metrics, specific company research. Either tool handles that. The rest is execution.