The short version. Ashby is the ATS to buy if you are scaling and want the AI already inside the system of record. Workable is the same decision for a smaller company at a lower entry price. Juicebox is the sourcing tool that does not need a sales call. Metaview handles interview intelligence. ChatGPT or Claude writes the job description, the outreach and the scorecard. And none of that is regulated, which matters enormously, because the moment any of these tools ranks or screens a candidate you may have acquired an annual audit obligation you do not know about.
Prices below were read off each vendor’s own pricing page on 15 August 2026, except the assistant tier. OpenAI’s page blocked our request, so we describe rather than quote it. Recruiting software is unusually good about publishing prices compared with the legal and accounting categories, which is worth knowing before you accept a sales call.
Own hiring rather than run the requisitions? Our directory keeps shortlists for people and hiring leads, covering interview structure and salary bands.
How we ranked, when the tool is regulated
Recruiting is the one profession on this site where the AI is not just a productivity tool. It is, in several jurisdictions, a regulated decision system. That reshapes the ranking, because a tool that saves an hour a week and creates a compliance obligation is a different purchase from one that saves an hour a week.
So we weighted:
- Does it decide, or does it draft? A tool that writes your job posting carries no regulatory weight. A tool that scores, ranks or filters candidates may make you the operator of an automated employment decision tool.
- Is it inside the system of record? Recruiting data fragmented across four tools is how candidates fall out of process and how you fail to answer an audit later.
- Can you price it without a sales cycle? Most of this market quotes by demo. The ones that publish are doing you a favour and we have weighted it.
- Does it work at your headcount? This category prices by company size, so the right answer at 30 employees is the wrong one at 600.
Read the compliance section below before you buy anything here. This is not boilerplate. It is the part of this list most likely to cost you money.
1. Ashby: best all-in-one if you are scaling
Foundations $400/month for up to 100 employees, including 1,500 AI credits a month. Plus (101–1,000 employees) and Enterprise are custom-priced, with 2,500 and 12,500 AI credits per seat per year respectively. Verified 15 August 2026.
Ashby’s argument is that ATS, CRM, scheduling and analytics belong in one system, and the AI features follow from that: candidate search, outreach personalisation, application review, talent rediscovery and interview summaries all operate on data the platform already holds.
That integration is the actual value and it is not really an AI story. Bolt-on AI tools re-derive context you already have, who applied before, which source produced hires, what the interview panel said, and get it slightly wrong. Talent rediscovery is the clearest case: surfacing a candidate who nearly got an offer eighteen months ago is only possible if the system holds that history.
Publishing $400/month up to 100 employees is also unusual in this market and worth rewarding. It is genuinely comparable before a demo.
Who it’s wrong for: companies hiring a handful of people a year, who are buying a talent platform for a spreadsheet problem. And organisations already standardised on Greenhouse or Lever. An ATS migration is a data and process project, not a software swap, and AI features are rarely worth it.
2. Workable: best ATS entry point for a smaller company
Standard $299/month ($3,588/year). Premier $599/month ($7,188/year). Enterprise $719/month ($8,628/year). All tiers list unlimited active jobs under fair use, and the Workable Agent add-on includes 3,000 AI credits, with candidate actions costing 1–10 credits depending on whether it is an evaluation, a sourcing action or a chat interaction. Verified 15 August 2026.
Two ATSs on a five-item list breaks a rule we normally hold to, and we are breaking it for the same reason we did on our legal list: these are mutually exclusive purchases and which one is right depends entirely on your size. Workable starts lower and reaches smaller companies; Ashby is built for the scaling case. Nobody buys both.
What to actually watch is the credit model rather than the headline price. Candidate actions consume 1–10 credits each, so a high-volume role, retail, hospitality, support, burns an allowance at a completely different rate than three engineering hires. Model your real requisition volume against the credits before choosing a tier, because that is where the true cost of this plan lives.
Who it’s wrong for: high-volume hourly hiring at scale, where conversational screening platforms are purpose-built and the credit arithmetic here turns against you. Also companies wanting deep analytics, which is where Ashby is stronger.
3. Juicebox: best sourcing without a sales call
Free tier with limited searches. Starter $99/month with 500 contact credits and 500 export credits. Growth $179/month with 1,500 contact and export credits per seat. Business custom. The Agents add-on is $199 per agent per month. Annual billing is 15% off. Verified 15 August 2026.
Sourcing is where recruiting AI is least controversial and most immediately useful, because finding people is not deciding about them. Natural-language search, describing the person you want rather than assembling a boolean string, is a genuine change in how the work feels, and it surfaces candidates a keyword search misses because their profile uses different words for the same experience.
Juicebox earns its place partly on price transparency. Its established competitors are sold by demo and quote, which makes them impossible to compare honestly in a list like this. At $99 a month a solo recruiter or small team can simply try it.
Watch contact credits, not searches. Searching is unlimited on paid tiers; reaching people is what is metered, and reaching people is the job.
Who it’s wrong for: in-house teams whose pipeline is inbound and referral. You would be buying outbound capability for a problem you do not have. And anyone expecting the agent add-on to run outreach unattended, which is both a quality risk and, if it starts ranking who is worth contacting, a compliance question.
4. Metaview: best for what happens inside the interview
Free tier covering 100 profiles sourced. Pro $100 per user per month for 200 profiles sourced monthly. Max $300 per user per month, unlimited. Enterprise custom. Note that its pricing page meters by sourced profiles. Verified 15 August 2026.
Interview notes are the most reliably bad artifact in recruiting. An interviewer types while listening, writes up an hour later from memory, and produces a scorecard that reflects recall rather than the conversation. Structured hiring depends on the record being accurate, and it usually is not.
Capturing the interview and producing structured notes against the scorecard fixes a real quality problem, and it does it in the place where hiring decisions are actually made rather than at the top of the funnel.
Recording interviews carries obligations of its own. Consent rules vary by jurisdiction and are not optional; some US states require all-party consent, and Illinois has specific law governing AI analysis of video interviews. Tell candidates, get consent, and check whether your tooling analyses the recording or merely transcribes it, those are different regulatory objects.
Who it’s wrong for: teams that will not change how they interview. Notes against a scorecard are worth little without a structured process to score against. Also anyone whose legal team has not signed off on recording candidates.
5. ChatGPT or Claude: best for the writing, and the only unregulated part
Both sell an individual tier around $20 a month. We describe rather than quote: OpenAI’s pricing page returned a 403 to us on 15 August 2026.
Job descriptions that are not copies of the last one. Outreach that references something real about the person. Turning a hiring manager’s rambling intake call into a scorecard with actual criteria. Rewriting a rejection so it reads like a human sent it. Interview questions that probe a competency rather than testing recall.
The intake-to-scorecard use is the underrated one. Most bad hiring processes fail at the start, when nobody wrote down what “good” means, and a structured conversation with a model is a cheap way to force that definition into existence before the first CV arrives.
Note what this section does not include: any use where the model reads candidates and tells you who is better. That single distinction is the difference between a $20 tool and a regulated decision system, and it is why this ranks fifth on utility and first on safety.
Who it’s wrong for: anyone who will paste a shortlist in and ask it to rank. Also check your data handling. Candidate CVs are personal data, and consumer, business and API tiers treat it differently.
The compliance section, which is the point of this list
If a tool substantially assists or replaces a screening decision, it is likely an automated employment decision tool, and the obligation attaches to you, the employer, not to the vendor who sold it.
New York City, Local Law 144. In force since 2023. Covered employers and agencies must obtain an annual independent bias audit of any AEDT used to screen candidates for employment or promotion, publish a summary showing selection rates and impact ratios by race/ethnicity and sex including intersectional categories, and give candidates at least ten business days’ notice before use. Academic study of actual compliance has found published audit summaries to be very rare relative to the number of employers plausibly covered. Which tells you both that enforcement has been light and that the exposure is widespread.
The EU. Employment and worker management are classified as high-risk uses under the AI Act. Separately, Article 50’s transparency obligations became applicable on 2 August 2026, which reaches AI systems interacting directly with people. A screening chatbot talking to candidates is squarely in scope.
Illinois regulates AI analysis of video interviews specifically, including notice, explanation and consent. Other states have moved or are moving; Colorado’s AI legislation covers consequential decisions including employment.
The practical version for a recruiting team: know which of your tools score or rank rather than draft or find, ask every vendor in that first group for their bias audit and what it covers, and remember that their audit of the model is not your audit of your use of it. The cheapest position is to keep AI on the drafting and sourcing side of the line and keep humans on the deciding side. Which, conveniently, is also where the tools are actually good.
Recruiting teams this won’t fit
High-volume hourly hiring. Thousands of applicants a month is a different product category built around conversational screening and automated scheduling, and it is also the deepest end of the compliance pool.
Executive search. Your work is relationships and market mapping. Sourcing automation does not help, and the candidates would rather you did not.
Agencies billing on placement. Your economics are different enough that per-seat tooling maths does not transfer.
Anyone hoping AI fixes a hiring problem. Most broken hiring is an undefined role, a slow process and an indecisive panel. AI makes an undefined role fill faster with people nobody agreed on.
Deliberately excluded
Greenhouse, Lever and the enterprise ATSs. Capable and sold by quote. We cannot compare them honestly against products that publish prices, and readers at this size can get a demo without our help.
Paradox, HireVue and the screening platforms. Deliberately excluded. They are the products most likely to be AEDTs, they are sold by sales call, and recommending one in a general list without knowing your jurisdiction, volume and audit posture would be irresponsible. If you need conversational screening at volume, that is a procurement project with legal in the room.
LinkedIn Recruiter. Effectively unavoidable for many teams and priced by negotiation. Not a decision most people are making freely.
Bias-audit platforms. A real and growing category, and the right answer if you already operate an AEDT. Left out because they are engagement-priced and because they are a consequence of a purchase rather than a purchase in their own right. You buy one because you bought something else.
Recruiter questions
Does using AI to screen resumes require a bias audit?
In New York City, if the tool substantially assists or replaces the screening decision, yes. An annual independent bias audit, a published summary of selection rates and impact ratios, and ten business days’ notice to candidates. Other jurisdictions have their own rules. Drafting a job description or sourcing a name does not trigger this; scoring or ranking candidates can.
Is my vendor’s bias audit enough?
Generally no. A vendor’s audit covers their model; the obligation covers your use of it, on your candidate pool, for your roles. Ask what the audit measured, on whose data, and when. Then ask your own counsel whether it satisfies your obligation. Assuming it does is the most common mistake in this area.
Which of these five carry compliance risk?
The risk follows the function, not the brand. Sourcing and drafting are low-risk. Any feature that scores, ranks, filters or recommends candidates is the one to examine. And several of the ATSs here include such features that you can choose whether to enable. Find out what is switched on.
What should a small company buy first?
An ATS, before any AI. If candidates live in an inbox you have a process problem that no model fixes, and you cannot answer a compliance question about a process you cannot reconstruct. Workable’s Standard tier at $299/month is the realistic floor here; add sourcing when your inbound pipeline genuinely runs dry.
Can I record interviews and have AI take notes?
Usually, with consent, and the rules vary. Some US states require all-party consent, and Illinois specifically regulates AI analysis of video interviews with notice and explanation requirements. Also distinguish transcription from analysis. A tool that evaluates a candidate from the recording is a different regulatory object from one that writes down what was said.
Will AI replace recruiters?
It is absorbing sourcing, scheduling and first-draft writing, genuinely a large share of the hours. What it cannot do is persuade a strong candidate to take a lower offer, manage a hiring manager who keeps changing the brief, or carry the legal responsibility for a screening decision. The exposure is to coordination work, not to the judgment or the accountability.
See our maintained directory picks for people and hiring leads, including pay transparency checks on live job posts.
Last reviewed: 15 August 2026. Prices read off vendor pricing pages on that date, except the assistant tier, which we describe because OpenAI’s page blocked our request. Nothing here is legal advice; employment and AI regulation vary by jurisdiction and move quickly, and your counsel governs. We re-review quarterly.
Leave a Reply