To choose an ATS in 2026, score every shortlisted system against two layers at once: the baseline features that run daily hiring, and the AI-era criteria that most older checklists still skip. Weight each criterion, test it live in a demo, and keep only the systems that clear both.

Most 2026 buying decisions come down to the AI-era layer. AI screening has shifted from a nice extra to a baseline expectation, and the controls sitting behind it now carry real legal weight and real hiring-quality weight. Throughout, Sprad serves as a working example of a system that already covers those criteria.

A strong shortlist comes down to a handful of decisions that are easy to get wrong under sales pressure.

What baseline features must every ATS still cover?

Before any AI conversation, an ATS has to nail the fundamentals, because a broken basic quietly costs you more hires than a missing AI feature. Six capabilities form the non-negotiable baseline that runs your hiring day to day.

Start with the career page. Around 60% of candidates abandon an application that runs past 15 minutes, so the length of your form directly decides how many people finish it, and shortening it is the fastest fix you have. Response discipline matters just as much, because 53% of job seekers say an employer has ghosted them, most often right after they applied, while a genuinely positive experience pushed 66% of candidates to accept an offer.

Integration depth has quietly become a baseline criterion in its own right. The average organization now runs about 16 separate HR and recruiting applications, and 68% of teams operate on disconnected platforms, so you end up re-keying data by hand from week one whenever your ATS cannot talk cleanly to the rest of your stack.

Which AI-era criteria do most ATS checklists miss?

Five criteria separate a 2026-ready ATS from an older one, and older checklists barely mention them. They run from the quality of AI screening all the way to how well a system mines the talent pool you already own. AI is not optional anymore. Use across HR tasks hit 43% in 2025 in mostly North-American surveys, and recruiting was the top use case. Gartner expects nearly every recruiting-tech vendor to have AI built in by 2027.

AI screening quality and explainable scoring

Screening quality is only part of it. The bigger stumbling block is whether you can actually explain a given score. A University of Washington audit of AI resume screening found language models favored white-associated names 85% of the time and never preferred Black-male-associated names over white-male ones, so a scoring model you cannot inspect becomes a liability the moment a rejected candidate asks why. Before you shortlist anyone, make them show you the evidence behind a score. Sprad screens bias-blind against the job criteria and ties an evidence quote to every rating, with a human always making the final call.

Open API and MCP control

You keep control of your own data only while the integration layer stays open. The Model Context Protocol, an open standard often called the USB-C for AI, lets any AI agent connect to any compatible system through a single interface, which means the old promise that a vendor integrates with your ATS has stopped being a real differentiator. What matters now is what a system actually does with the data once it is connected: does the scoring hold up, and does the output stay structured and compliant? Gartner expects 40% of enterprise apps to embed task-specific AI agents by the end of 2026, so an open API with native MCP support keeps you free to connect your own AI, such as Claude or ChatGPT, to the pipeline. Hardly any applicant tracking systems offer that combination. Sprad builds it into its free core.

Good to know: native MCP support is still uncommon in 2026, so if keeping your own AI in the loop matters, confirm it during the demo rather than assuming an open API alone is enough.

Bias auditing and EU data residency

An AI ATS can quietly turn into legal exposure if the compliance controls are missing. The EU AI Act classifies recruitment and selection AI as high-risk, which puts bias testing and documented human oversight on the companies that deploy it, with fines reaching 15 million euros or 3% of global turnover. Those high-risk obligations now apply from 2 December 2027 after the Digital Omnibus deferral. New York City already demands an independent bias audit of automated hiring tools, and no vendor may audit its own system. Because an ATS processes candidate personal data, GDPR adds a lawful basis and deletion workflows on top, and EU data residency makes the whole picture easier to govern. Sprad hosts candidate data in the EU, keeps a full audit trail of every AI action and leaves the final decision with a person. Get that right and you can actually defend yourself under both the EU AI Act and GDPR.

Talent-pool intelligence

Talent-pool intelligence is the most underused criterion on this list. Vendor figures suggest roughly 75% of candidates in an ATS are never contacted again after their first application, even though rediscovered candidates convert to hires around three to four times faster than cold-sourced ones, and about 44% of strong hires are already in the database. Those numbers are vendor-originated and directional, yet the logic holds: a searchable, consent-managed pool turns every past applicant into a warm lead for the next role. Sprad's Living Talent Pool lets you search past applicants in plain language and keeps consent and deletion handling built in.

How do you score an ATS with a weighted checklist?

Score the criteria so a good demo cannot paper over a weak system. Give each criterion a weight, rate every shortlisted ATS from 1 to 5 during the demo, multiply weight by score, and total the columns. The template below sums to 100%, so tune the weights to your own hiring reality before you start scoring.

The weighting stops one impressive feature from carrying the whole decision. A system can look strong on baseline features and still lose the deal on a single AI-era row, such as a black-box screening model or missing EU hosting, and a weighted total makes that trade-off visible at a glance.

Which ATS red flags should stop a purchase?

Three red flags cost buyers the most after signing, and each hides easily inside a polished demo. It is a short list, but run through it line by line before you sign anything.

Test before you sign: ask for a full data export during the trial and confirm exactly what comes out, in which format, and whether candidate history, audit trail and attachments come with it.

Per-seat pricing deserves a closer look, because the sticker price rarely equals the real bill. Published ATS pricing runs from $0 to well past $125,000 a year, and setup fees, seat expansion and multi-year prepayment routinely push total cost of ownership 30 to 50% above the base subscription. Our breakdown of what vendors actually charge maps those bands to company size before you negotiate.

How do the criteria change with company size and hiring volume?

The criteria stay the same across company sizes, but the weights move, and hiring volume matters as much as headcount. A five-person startup and a 5,000-person enterprise both need explainable AI and clean data handling, yet they load the scoring sheet very differently.

For small teams and low hiring volume, a genuinely free-forever recruiting core usually covers the job, so the priority is simply avoiding a per-seat plan that punishes you for growing.

For mid-market teams between 50 and 500 employees, the fastest-growing slice of the applicant-tracking category, AI screening and integration depth are what pay off first, and this is where per-seat creep does the most damage to a budget.

Large enterprises account for roughly 67% of ATS revenue and hire across many locations, so they weight data residency and open API control most heavily, with budgets that stretch past $125,000 a year and six-figure setup projects. Volume changes the math too: at a US benchmark cost-per-hire of around $5,500, shaving even a little time off screening adds up fast, though that figure runs high for DACH.

The layer that actually decides your shortlist

The baseline features are now so widely matched that nearly every credible ATS passes them, which is why they should carry less of your score than instinct suggests. Your real decision, and almost all of your legal exposure, sits in the AI-era layer, so those rows deserve the heaviest weights.

So run the weighted sheet on two or three systems in a live demo, ask for the evidence behind one AI score, and test a full export before you sign. On the AI-era rows, Sprad is a useful benchmark: its free ATS core already carries EU hosting, an open API with native MCP, and a bias-blind, human-in-the-loop screening flow.

FAQ: choosing an ATS in 2026

What features should an ATS have?

A capable ATS covers two groups of features. The baseline group runs daily hiring: a fast career page, job multiposting, a candidate pipeline with team collaboration, and reporting that plugs into clean integrations. The 2026-ready group is what separates modern systems, adding explainable AI screening, an open API with MCP, independent bias controls and EU data residency. Weight that second group heavily when you compare.

What should I look for in an AI ATS?

In an AI ATS, the first thing to check is what the AI actually does with candidate data once it is connected. Prioritize scoring you can explain against clear job criteria, plus an open API with MCP so you keep control of your own AI. Independent bias testing and EU data residency cover GDPR and the EU AI Act. You also want a visible audit trail and a human making the final call.

What is ATS export lock-in?

Export lock-in means your data comes out only as a flat file, while candidate history, audit trails, attachments and record relationships stay trapped in the old system. It makes switching vendors slow and costly. Avoid it by testing a full export during the trial and confirming exactly what leaves the platform and in which format.

How much should an ATS cost?

Anywhere from $0 to more than $125,000 a year. Small businesses typically pay $250 to $3,000, firms with 100 to 500 employees $3,000 to $15,000, and large enterprises $125,000 or more, with per-user plans often $50 to $150 per user each month. Budget for setup fees and per-seat expansion on top of the sticker price.

Should a small company pay for an ATS?

Not necessarily. A small company can run real hiring on a genuinely free-forever core that includes a career page, multiposting and a candidate pipeline, then pay only for AI modules once volume justifies it. Paying makes sense when deep integrations, per-seat collaboration or high screening volume start saving more hours than they cost.