TL;DR: An AI interviewer RFP should cover ten things: scope and volumes, scoring, human oversight, candidate experience, ATS and workflow, compliance, security, implementation, pricing units and proof. Copy the questions below, send them to every vendor, then score the answers with the weighted scorecard at the end.
Updated 18 November 2026. ZeaHire is our product, so it appears once, near the end.
When you need this
Use it when you are about to put an AI interviewer in front of candidates and want written answers before you sign. It suits a staffing owner or talent acquisition lead with two to five vendors on a shortlist. Send it after a first demo and before pricing talks. Paste the first column into your RFP and keep the other two for scoring.
1. Scope, volumes and interview formats
Every later answer depends on this one. A product tuned for a few dozen interviews a month can struggle with thousands, and vendors will say yes to a loose brief. Give each one your real role types, monthly volumes and busiest weeks. Listen for named role categories, the largest load they run today, and plain limits on phones and languages. "We support any role" is not an answer.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Which roles and volumes is the product built for? | Fit at 50 interviews a month is not fit at 5,000. | Named role types and the largest customer volume it handles. |
| What interview formats do you offer: voice, video, chat or text? | The format changes who finishes the interview. | Each format, and when to use which. |
| Can candidates take the interview on a phone? | Many candidates apply from a phone. | Browser or app, any install needed, device limits. |
| Which languages can the interview run in? | A translated menu is not a language interview. | Languages for questions, answers and the scorecard. |
2. Scoring and how it is explained
The score is what recruiters pass to clients, so this section decides whether you can defend a shortlist. Trouble starts when a vendor shows one number with no evidence, or when the tool decides what a good answer sounds like. Ask to see a full scorecard from a real interview. Listen for who sets the standard (it should be you) and whether you hear before scoring changes. Vague reasoning now means arguments with clients later.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| What does the scorecard contain? | A bare number cannot be defended. | Overall score, dimensions, evidence and a transcript. |
| Does every candidate get the same questions? | Consistent questions make scores comparable. | One template per job, applied to everyone. |
| Who defines what a strong answer looks like? | Your standard should drive the score, not the vendor's. | Recruiter-set criteria per question. |
| Can you show the reasoning behind each score? | Clients and candidates will ask why. | Written reasoning per question. See explainable AI hiring. |
| What happens when the model or scoring changes? | Silent changes break comparisons. | Notice of changes, versioning, and a way to compare. |
3. Human oversight
An AI interviewer should speed up the first conversation, not make the hiring decision. The common failure is a quiet default: auto-advance or auto-reject unless you find the setting, or a shortlist sent onward before a recruiter looks. Ask how each path works today, not how it could work. A strong vendor names the approval steps, shows the override log and says what the system never does alone. Be wary of "assist" with no named person.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Who makes the final decision on each candidate? | Your team stays accountable. | A named person approves, rejects and advances. |
| Can the AI reject anyone on its own? | Auto-rejection raises legal and fairness risk. | A clear "no", or a setting you control. |
| Can a recruiter override a score, and is it logged? | Overrides are your quality control. | Override with a reason, kept in an audit trail. |
| What goes to a client or hiring manager, and who sends it? | Nothing should leave without your review. | A manual submit step, logged per job. |
4. Candidate experience and accessibility
Candidates decide whether your pipeline works. A confusing or glitchy interview loses good people, and notice, consent and accommodation duties land on you. Vendors often show a polished recruiter dashboard and skip the candidate side, so take the interview yourself, on your phone. Listen for the actual notice text, a real alternative for people who opt out, and accessibility evidence you can test, not a general promise to be inclusive.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| What does the candidate see before, during and after? | Surprises cost completions. | A walkthrough you can take yourself. |
| How is the candidate told AI is involved, and how do they consent? | Notice rules vary by place. | The exact text shown and a time-stamped record. |
| What is the alternative path for someone who cannot or will not take an AI interview? | Accommodations are a legal and ethical need. | A human-led option and a way to request it. |
| Which accessibility standard does the candidate flow meet? | You need something testable. | A named standard (such as WCAG 2.2, read 18 Nov 2026) and test evidence. |
5. ATS and workflow
An interviewer outside your ATS means double entry, and double entry kills adoption. "Integration" covers everything from a one-way link to a deep connection, and vendors use the word freely. Name your exact systems and ask what moves, in which direction and when, shown with your own account. A strong answer separates what is live from what is planned, gives a firm timeline for anything built for you, and explains how duplicates are handled.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Which ATS and HR systems do you connect with today? | "Integration" can mean a lot of things. | A list, and which are live versus on request. |
| Will you build a connection to our system if you lack one? | Your stack may be unusual. | A yes or no, scope and a timeline. |
| What goes back to our ATS, and when? | Scores, notes and stage changes differ. | Field-by-field detail, shown with your own account. |
| How are duplicate candidates handled? | Two records waste recruiter time. | The matching rule and how conflicts surface. |
| Who can see what inside the product? | Candidate data should be limited. | Role-based access you can edit. |
More on this in AI interview ATS integrations.
6. Compliance
Hiring-AI rules vary by city, state and country, and several are still changing. The aim is not a verdict from the vendor but to learn who carries which duty, theirs or yours, and whether they can hand you paperwork. The usual problem is a confident "we're covered" with nothing behind it. Ask for documents, and listen for named jurisdictions, an identifiable auditor and a process for telling customers when rules shift. Law-by-law detail is on the compliance questions page. Rules change, so check with your legal adviser.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Which hiring-AI laws apply to your product, and where? | Duties differ by city, state and country. | A jurisdiction list and who carries which duty, vendor or customer. |
| Can you export the records we may need to produce? | Retention rules differ by place. | Per-candidate inputs, outputs and decision history. |
| Do you provide a bias audit or adverse-impact summary, and notice text? | Some laws ask; buyers should anyway. | Method, independent auditor, date, shareable summary, editable notice. |
| How do you track changing rules? | Colorado, for example, replaced its 2024 law with SB 26-189, signed 14 May 2026 (read 18 Nov 2026). | A named owner and a customer notification process. |
7. Security and privacy
Interviews produce sensitive material: recordings, transcripts, scores and sometimes demographic data. A common gap is a security page full of badges that describe a different system than the one you are buying. Ask for documents, and check what each one actually covers. Listen for contract language on model training, a current sub-processor list, and written retention and breach terms. The full list is on the security questionnaire page.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Which candidate data do you collect and keep? | You cannot protect what you cannot list. | Video, audio, transcripts, scores, demographics, each with a purpose. |
| Do you hold an independent security report, such as a SOC 2 Type II report or ISO/IEC 27001 certificate? | A claim is not evidence. | The report, its period or scope and the system it covers. |
| Does any of our data train your models or anyone else's? | Candidate data is sensitive. | A contract clause, not a policy page. |
| Who are your sub-processors, and where do they process data? | Your data travels through them. | A current list with function and location, and change notice. |
| What are your deletion, retention and breach-notice terms? | You carry the risk to candidates. | Defaults, how to change them, written proof of deletion and a notice window. |
8. Implementation and support
A good product still fails when rollout stalls. Vendors describe their own steps and leave out yours, such as who approves templates, loads users or maps fields, and they rarely volunteer how to leave. Ask for a plan with named owners on both sides. Listen for a realistic timeline, training for recruiters as well as admins, and exit terms covering record export and deletion. If the answer is "just turn it on," press for detail.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| What are the steps from signature to first live interview? | Rollout drags without a plan. | Named steps, owners and your own tasks. |
| What do you need from us during setup? | Your time is a cost. | Roles, data and approvals. |
| What does training and support look like? | Adoption decides value. | Training for recruiters and admins, and support channels. |
| What are the exit terms? | You may want out. | Export of records, scores and audit trails, plus deletion on exit. |
9. Pricing model
Totals mislead until every vendor prices the same thing. A per-interview, per-candidate, per-seat, per-job or flat platform fee each moves differently as volume changes, so ask for the unit first and figures second. Then run your own volumes through each model, including a slow month and a spike. What goes wrong is the add-on: connections, languages or support priced separately. Listen for one defined unit and an itemized list of extras, and compare the cost of a completed screen.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| What is the unit: per interview, per candidate, per job, per seat or a platform fee? | Units behave differently as volume moves. | One unit, defined, with what counts toward it. |
| What happens when volume spikes or drops? | Seasonal and per-requisition work is uneven. | Tiers, overage rules and how you can scale down. |
| Which items cost extra: ATS connections, languages, support? | Add-ons change the total. | An itemized list. |
| What are the term and renewal terms? | Lock-in affects leverage. | Term length, notice and renewal changes. |
10. Proof: pilot and references
A demo shows a product on a good day with a prepared script. A pilot shows it with your roles, candidates and recruiters. The usual failure is a pilot with no agreed measures, so everyone judges by instinct. Ask what the vendor supplies, how long it runs and what you will both count. For references, listen for customers like you in size and type, and for a candid story about something that went wrong.
| Question | Why it matters | A strong answer includes |
|---|---|---|
| Can we run a pilot on one role before we commit? | A demo is not your data. | Scope, length and what the vendor supplies. |
| What success measures do you propose? | You need numbers before you start. | Completion rate, recruiter time, submittal speed and client acceptance. |
| Who can we call as a reference, in a business like ours? | Fit is specific. | Two or three customers of similar size and type. |
| What went wrong for a past customer? | Honest answers show maturity. | A specific story and the fix. |
How to score vendor answers
Score each section 0 to 5, scale by weight, and total out of 100. Change the weights to match your risk.
| Section | Weight | What earns full marks |
|---|---|---|
| Scoring and explanation | 15 | Same questions for all, reasoning per answer, your own criteria. |
| ATS and workflow | 15 | Your system listed or a firm build plan, with field-level detail. |
| Scope, volumes, formats | 10 | Proven at your volume, phone and language fit. |
| Human oversight | 10 | A person decides, overrides are logged, nothing auto-sent. |
| Candidate experience | 10 | Walkthrough, notice, alternative path, accessibility evidence. |
| Compliance | 10 | Named duties, audit summary, exportable records. |
| Security and privacy | 10 | Reports shown, no training on your data, clear sub-processors. |
| Proof: pilot and references | 10 | Pilot offered, measures agreed, comparable references. |
| Implementation and support | 5 | Named plan, clear exit terms. |
| Pricing model | 5 | One clear unit, itemized extras. |
Two rules keep it honest: an answer scores once it is shown or documented, so a "we'll show you in a demo" counts after you have seen it with your own account; and you score answers before you see prices. For the wider choice, see which AI recruiting platform to buy.
RFP timeline (illustrative)
Adjust to your own approvals.
| Week | Step |
|---|---|
| 1 | Define roles, volumes and must-haves. |
| 2 | Send the RFP to two to five vendors. |
| 3 | Vendors return answers and documents. |
| 4 | Score answers; demos with your own ATS. |
| 5 | References, security and legal review. |
| 6 | Agree pilot scope and measures. |
| 7 to 10 | Run the pilot on one role, then decide. |
Mistakes to avoid
- Asking for features, not evidence. "Do you have bias audits?" gets a yes. "Show the summary and auditor" gets the truth.
- Skipping volumes. Vendors cannot judge fit or price without them.
- Treating compliance as the vendor's job. Ask who owns which duty, in writing.
- Judging the demo, not the answers. Polish is not proof.
- Settling on a total before the pricing unit. Totals are not comparable until every vendor prices the same unit.
- Skipping the pilot. A demo is not your data; run one role before you commit.
How ZeaHire answers these
ZeaHire is our product. Here is how it answers some of the questions above.
- Screening: ZeaHire scores each resume against the job, with skills, experience and education sub-scores plus matched and missing skills. A person approves or rejects each new application before it enters the pool.
- Scoring: every candidate on a job gets the same template questions. Recruiters set what a strong answer covers, and each answer is scored against it with written reasoning. The scorecard has an overall score and recommendation, five skill dimensions with evidence, per-question reasoning and a transcript.
- AI-led interviews: the candidate takes the interview from an emailed link, with no recruiter on the call. Length is 10 to 60 minutes, in English. A human panel can run an interview in ZeaHire or Microsoft Teams, with an AI note-taker that suggests follow-ups.
- Human decision: recruiters approve, reject, move stage and submit to the client; nothing goes to a client on its own, and submissions are logged per job.
- Candidates and integrity: candidates can rate the interview. Proctoring signals are recorded in each AI-led interview, and candidates verify with a code sent to their invite email.
- Access and agencies: role-based access with an editable permission matrix, and Sign in with Microsoft. A user can work across more than one organization, and jobs or candidates can move between them.
- ATS: ZeaHire connects with nearly a dozen ATS and HR systems, including Workday, SAP SuccessFactors, Greenhouse, JazzHR and JobDiva. If yours isn't on the list, we'll build the connection. Ask to see it with your own system in a demo.
For compliance, security and pricing rows, ask us for ZeaHire's answers to this list in your demo.
FAQ
What should an AI interviewer RFP include?
Scope and volumes, interview formats, scoring and how it is explained, human oversight, candidate experience, ATS and workflow, compliance, security and privacy, implementation, pricing units, and proof through a pilot and references. Add a scorecard so every vendor is rated the same way.
Should an RFP include a budget?
It is your call. Volumes matter more, because vendors cannot size a fit without them. A range may tighten proposals; holding it back keeps pricing units easier to compare. Either way, ask for the unit first and figures second.
Can AI conduct interviews?
Yes. An AI interviewer can ask a job's structured questions and record the answers, and the candidate often takes it from a link with no recruiter on the call. A person should still decide who moves forward. See how AI interviews work.
Should AI be used in the hiring process?
That depends on your roles and the controls around it. Use it where structure helps, keep a person accountable for decisions, and ask for evidence on scoring, fairness and data handling first.
How do you write an RFP for software?
State the problem, your volumes and your systems. Group the questions, set a deadline and a scoring method before vendors reply, and ask for documents, not yes or no answers.
Is an RFP a contract?
No. It is a request for proposals. The contract comes later and should carry the answers you accepted.
Go deeper
- Agencies: what to add for several clients and fast submittals.
- Compliance: law by law vendor questions.
- Security: the security questionnaire.
Book a demo and bring this list. We'll walk through our answers with your own ATS.
Comments
No comments yet — start the conversation.