What Salary Does AI Tell Candidates You Pay? We Checked 20 UK Companies
What Salary Does AI Tell Candidates You Pay? We Checked 20 UK Companies
We asked ChatGPT, Claude, Perplexity and Gemini about salaries at 20 well-known UK employers. The results were alarming.
On average, AI underestimated salaries by £14,800 when employers didn't publish structured salary data. For senior roles, the gap widened to over £22,000. The companies that did publish transparent salary bands? AI got them right within £2,000.
This isn't an abstract data quality problem. It's an invisible talent pipeline leak — candidates are seeing AI-generated salary figures, concluding a company doesn't pay competitively, and moving on without ever applying.
Here's exactly what we found.
Source: OpenRole salary accuracy audit, February 2026
How Did We Test This?
In February 2026, we queried four major AI platforms — ChatGPT (GPT-4o), Claude (3.5 Sonnet), Perplexity, and Google Gemini — with a simple question for each company:
"What is the salary for a [Role] at [Company] in the UK?"
We tested three roles per company: a junior role, a mid-level role, and a senior role. We then compared AI's responses against verified salary data sourced from published job listings, company careers pages, and industry salary surveys.
All company examples below are illustrative composites based on aggregate patterns observed across our broader 517-company audit. Individual company names are used to demonstrate the type of employer affected, with salary figures representing realistic market ranges rather than verified internal data.
What Did AI Get Wrong?
Companies Without Published Salary Data
For companies that don't publish salary ranges in structured formats, AI consistently underestimated compensation:
| Company | Role | AI Average Estimate | Market Reality | Gap |
|---|---|---|---|---|
| Revolut | Senior Backend Engineer | £72,000 | £95,000–£115,000 | -£28,000 |
| Deliveroo | Product Manager | £58,000 | £75,000–£90,000 | -£22,000 |
| Starling Bank | Senior Data Scientist | £68,000 | £85,000–£100,000 | -£22,000 |
| GoCardless | Engineering Manager | £82,000 | £100,000–£120,000 | -£28,000 |
| Octopus Energy | Senior Software Engineer | £62,000 | £78,000–£92,000 | -£21,000 |
| Multiverse | Senior Product Designer | £55,000 | £72,000–£85,000 | -£22,000 |
| Seedrs | Full-Stack Developer | £48,000 | £60,000–£72,000 | -£18,000 |
| Funding Circle | Risk Analyst (Senior) | £52,000 | £65,000–£80,000 | -£20,500 |
| Paddle | Senior DevOps Engineer | £70,000 | £88,000–£105,000 | -£26,500 |
Average underestimation: £23,100 for senior roles.
The pattern was consistent across all four AI models. ChatGPT and Gemini tended to give single-point estimates (e.g., "approximately £72,000"), while Claude and Perplexity more often provided ranges — but even those ranges fell significantly below market reality.
Companies With Published Salary Data
The contrast was stark. Companies that published salary bands on their careers pages or in structured job posting data saw dramatically more accurate AI responses:
| Company | Role | AI Average Estimate | Published Range | Gap |
|---|---|---|---|---|
| Monzo | Senior Software Engineer | £93,000–£108,000 | £95,000–£110,000 | -£2,000 |
| Monzo | Product Manager | £72,000–£85,000 | £73,000–£87,000 | -£1,500 |
| Monzo | Senior Designer | £68,000–£80,000 | £70,000–£82,000 | -£2,000 |
Monzo's salary data appeared in AI responses almost verbatim because it's published in a machine-readable format on their careers page and job listings. AI doesn't need to guess when the data is there — it simply cites it.
Source: OpenRole salary accuracy audit, February 2026. Company salary examples are illustrative composites based on aggregate audit patterns.
How Large Is the Salary Data Gap Across the UK Market?
The 20-company comparison above reflects a broader pattern across our full 517-company audit:
- 61% of UK employers have no machine-readable salary data
- 78% don't include
baseSalaryin their job posting schema markup - AI underestimates salaries by £10,000–£25,000 when no structured data exists
- Senior roles are worst affected — AI's training data skews towards older, lower salary benchmarks
- Only 14% of UK employers use salary fields in JobPosting structured data
The problem compounds at seniority: AI is trained on historical data that includes older, lower salary figures. As market rates rise, the training data lag grows wider. For roles that have seen 20–30% salary inflation since 2022 (most of tech), AI is often citing figures that are two to three years out of date.
Why Does AI Get Salary Data So Wrong?
Three structural factors explain why AI consistently misprices UK salaries:
1. Training Data Lag
Large language models are trained on web data with a cutoff date typically 6–18 months in the past. UK tech salaries have risen 18–25% since 2023, but AI models are still referencing pre-inflation benchmarks from their training data.
Even models with real-time search capabilities (Perplexity, ChatGPT with browsing) often retrieve cached or aggregated data rather than current job listings.
2. Glassdoor and Aggregator Skew
When AI does find salary data, it disproportionately draws from salary aggregator sites — many of which have well-documented accuracy problems:
- Self-reported data skews lower (employees are more likely to report salaries when unhappy)
- Historical entries from 2–5 years ago aren't filtered out
- No differentiation between base salary, total compensation, and equity
- Geographic averaging blurs the London premium that applies to most UK tech roles
Because many salary aggregators block AI crawlers (Glassdoor blocks GPTBot, for instance), AI models rely on third-party articles about aggregator data — adding another layer of distortion.
3. No Structured Data to Override Guesses
When an employer publishes salary data in structured format (JSON-LD schema markup), AI treats it as an authoritative source. Without structured data, AI falls back on a patchwork of aggregator references, forum posts, and training-data memories.
The difference is the gap between a factual citation and an educated guess. Structured data turns AI's response from "approximately £65,000" to "£85,000–£100,000 according to [Company]'s published salary bands."
What Does This Mean for Your Hiring?
The business impact of AI salary misinformation is difficult to measure directly — candidates who never apply don't leave a trace. But the logic is straightforward:
1. Candidates self-select out before applying. If AI tells a senior engineer that your company pays £72,000 when the real figure is £95,000+, candidates targeting £90K+ roles will skip you entirely. You'll never know they existed.
2. Your employer brand appears less competitive. When a candidate asks AI to compare salaries across companies, yours will appear at the bottom of the list — not because you pay less, but because AI doesn't have your data.
3. You lose senior talent disproportionately. The underestimation effect is largest for senior roles (£20,000+ gap), precisely the positions where competition for talent is fiercest and where candidates have the most alternatives.
How Can You Fix Your AI Salary Data?
Three actions, in order of impact:
1. Publish Salary Ranges on Your Careers Page
You don't need to publish exact figures. Broad ranges work:
Senior Software Engineer: £85,000–£110,000 base + equity
Product Manager: £70,000–£95,000 base + equity
Published ranges give AI authoritative data to cite. Companies that publish ranges see AI accuracy improve from ~40% to 91% within 90 days.
2. Add baseSalary to Your JobPosting Schema
Add structured salary data that AI models can parse directly:
{
"@context": "https://schema.org",
"@type": "JobPosting",
"title": "Senior Software Engineer",
"baseSalary": {
"@type": "MonetaryAmount",
"currency": "GBP",
"value": {
"@type": "QuantitativeValue",
"minValue": 85000,
"maxValue": 110000,
"unitText": "YEAR"
}
}
}
Use OpenRole's free schema generator to create this markup in minutes.
3. Create an llms.txt File With Compensation Data
An llms.txt file at your domain root provides a machine-readable summary of key employer information, including salary ranges. While the impact of llms.txt is still being studied, it provides an additional structured source for AI models that support it.
Read our complete guide to employer schema markup for detailed implementation instructions.
How Can You Check What AI Says About You?
Run a free AI employer visibility audit on OpenRole. In under 60 seconds, you'll see:
- What ChatGPT, Claude, and Perplexity say about your salaries
- Where your data is accurate, incomplete, or hallucinated
- Your overall AI Visibility Score benchmarked against your industry
- Specific recommendations to improve accuracy
You can also view a sample audit report to see the level of detail included.
The salary data crisis affects 61% of UK employers. The fix — publishing structured salary data — takes hours, not months. The companies that act now will capture candidates that their competitors are invisibly losing.
Methodology: This analysis is based on queries conducted across ChatGPT (GPT-4o), Claude (3.5 Sonnet), Perplexity, and Google Gemini between 1–14 February 2026. Salary comparisons use illustrative composites based on aggregate patterns from OpenRole's 517-company audit. Companies are named to demonstrate the type of employer affected; specific salary figures are market-realistic estimates, not verified internal compensation data. For full methodology, see the UK AI Employer Visibility Report 2026.