Frequently asked questions
The questions HR, Talent and procurement teams ask first. For the full data and security detail, see the Trust and Security page.
The product
What is Debiaser.ai?
Debiaser.ai gives HR, Talent and People teams a fast, consultant-grade unconscious-bias review of workplace documents. You upload a job description, policy or people process; in minutes you get a report that quotes each problem phrase, explains who it may deter and why, and proposes a specific rewrite — plus a revised version of every document in its original format. It is the document review our consultants do, delivered by an AI we have constrained to do it the GDP way.
What kinds of documents can I review?
Job descriptions, HR policies, performance-management frameworks, manager guidance and workplace communications — as individual files or as a whole connected system. It accepts docx, pdf, xlsx, pptx or pasted text. You can add context so related documents are read together rather than in isolation.
Who is it for?
HR, Talent Acquisition and People-team professionals — typically at larger organisations reviewing documents before they go out. A second audience is the organisation admin who manages a shared credit pool and approves members' credit requests.
How is this different from Grammarly, Textio or an inclusive-language checker?
Those are live writing aids that nudge you as you draft, mostly for job adverts. Debiaser.ai is a document-review engine: it takes documents your organisation already has — JDs, policies, whole performance frameworks — and returns an evidenced report plus revised files. Different job, and broader than job ads. A good test is to run a document your current tool has already passed through it and compare.
Why not just use ChatGPT or another chatbot?
Three reasons. It is constrained to GDP's methodology — a general chatbot will happily suggest quotas or strip genuine requirements; this cannot. It is built for documents — you get your file back, revised, with formatting intact, not a wall of chat text. And it is a controlled data environment — quoted cost up front, seven-day deletion, no training on your content, and a DPA your procurement team can sign.
We already work with GDP's consultants — how is this different?
It is the same methodology, in a different format. Debiaser.ai is a fast, self-service first-pass review — minutes and about £10 per document, rather than a scoped engagement over days or weeks. It is ideal for reviewing documents at volume, sanity-checking a draft before it goes out, or triaging which materials need deeper attention. It complements consulting rather than replacing it: for complex programmes, sensitive change work or organisation-wide strategy, our consultants still do what software cannot.
Our approach: legality and fairness
Is this legal for US hiring? Our legal team is cautious about DEI tools.
It is designed for exactly that environment. Debiaser.ai never recommends preferential treatment, quotas or demographic targets — that is a hard constraint, not a setting. It widens the applicant pool by removing unnecessary barriers — inflated requirements, jargon, exclusionary phrasing — while preserving every genuine requirement. Removing unnecessary barriers is lawful, and good hiring, everywhere including the US. It provides professional guidance, not legal advice.
Does debiasing mean preferential treatment?
No. The product is built around one principle — wider pool, same bar. It widens access to the full qualified talent pool by removing unnecessary barriers, and it never touches a requirement the role genuinely needs. It will not suggest lowering standards or preferring any group.
What does it change, and what does it leave alone?
It rewrites gendered and coded phrasing, inflated experience demands, credential requirements without cause, and insider jargon. It preserves every genuine qualification, all safety and legal requirements, real physical demands (stated more accessibly), and your organisation's voice and format. The sample report deliberately shows what it did not change as well as what it did.
See a full sample report→Accuracy and method
How accurate is it? What if it flags something wrong?
Every finding must quote the exact source wording and explain its reasoning, so your team stays the judge — nothing is changed behind your back. It is a first-pass expert review in minutes, not an oracle: where it is uncertain, it says so, and you decide what to adopt before anything is published.
Does it make decisions about people?
No. The service provides advisory document review only. It does not make automated decisions about individuals, does not profile candidates or employees, and does not rank or score people. A human should review all recommendations before publication or action.
Which AI model does it use?
Reviews are produced with a frontier AI model — Anthropic's Claude — which we have spent months constraining to do this one job properly. The model receives no tools, no database access and no network access; submitted documents are framed as untrusted data, and every result must match a strict structured schema before it is saved.
Which languages does it support?
The methodology is language-aware for English today. Other languages are on the roadmap — if that is a blocker for your team, we would like to hear it, as it directly shapes what we build next.
Data, security and privacy
Where does our data go, and who sees it?
Application data and uploaded files are stored in the EU (Convex, on AWS in Ireland, eu-west-1). Document content is analysed by Anthropic's Claude in the United States under a Data Processing Agreement incorporating the EU Standard Contractual Clauses and the UK Addendum; API inputs and outputs are not used to train models. Payments, transactional email, bot protection and error diagnostics use a small number of named subprocessors, each under equivalent safeguards.
See the full subprocessor register→How long do you keep our documents?
Uploaded documents, extracted text, model outputs, reports and revised documents are deleted from active service storage seven days after a review completes. Failed reviews and abandoned drafts are covered by the same purge jobs. The deletion date is shown on every review. Only billing and ledger records are retained, where accounting and tax rules require it. (Anthropic may retain API data for up to 30 days for trust-and-safety purposes; it is never used for training.)
Do you train AI on our content?
No. Customer content is never used by Debiaser.ai to train AI models, and the model provider is contractually prohibited from training on API inputs and outputs.
How is the service secured?
Sign-in is passwordless, so no passwords are stored. All access to your content is mediated by authenticated server functions, file download URLs are short-lived and signed, traffic is encrypted in transit and provider storage is encrypted at rest, and rate limits protect sign-in, trial claims, uploads and payments. The analysis model is also hardened against prompt injection: it has no tools, database or network access, and a scope guard rejects clear non-workplace misuse.
Do you hold SOC 2 or ISO 27001?
Not yet. Debiaser.ai does not currently hold SOC 2 or ISO 27001 certification; both are on the roadmap as the product matures for larger enterprise deployments. In the meantime the Trust page sets out exactly how data is hosted, processed, retained and secured.
Read the Trust and Security overview→Can we sign a Data Processing Agreement?
Yes. Organisation admins can execute our standard DPA electronically from the organisation page — enter your particulars, review the completed agreement, sign, and GDP countersigns; both parties receive the executed copy. The standard terms and subprocessor register are published for procurement review, and a separately negotiated agreement can take precedence where agreed.
Pricing and credits
How much does a review cost?
You pay per review in credits, not per seat. One credit covers 1,000 words analysed, with a minimum of five credits per review, so a typical short job description is about £10 on the Starter rate. Credit packs are Starter (25 credits, £50), Team (100 credits, £180) and Volume (500 credits, £800). Credits never expire.
See full pricing→Are there free credits to try it?
Yes. Every account gets 10 trial credits on sign-up — enough to run a real document end to end before you buy anything.
What happens if a review fails?
Failed reviews are refunded automatically. You are only charged for reviews that complete.
Do you offer invoicing and purchase orders?
Yes, for organisations that need more than card checkout. We can invoice against a purchase order with volume pricing — no card needed — while your team keeps evaluating on trial credits so the process does not stall.
Do we pay per user?
No. There are no per-seat fees. An organisation buys credits once and grants allowances to individual team members from a shared pool.
Getting started and enterprise rollout
How do I start a review?
Sign in with your work email (passwordless), claim your trial credits, and upload a document or paste text. You will see the exact credit cost before anything is analysed; confirm it and the review runs in minutes.
Start a review→Can I see the output before uploading anything?
Yes. A full sample report on a fictional job description is published with no sign-up required, so you can judge the depth and tone of the output first.
Open the sample report→Can we get a walkthrough?
Yes. We are happy to walk you and your team through a live review in about 20 minutes and, if you have a document handy, run it together on trial credits.
Book a walkthrough→What does enterprise rollout look like?
Organisations get a shared credit pool with member allowances, a self-serve e-signed DPA and published subprocessor register, invoicing and purchase-order support, and volume pricing for teams reviewing high document volumes. GDP can join procurement, legal and IT-security conversations directly.
Still have a question?
Run a document on trial credits, read a full sample report, or talk it through with us.