If you've applied through a jobs.ashbyhq.com link, a model may have read your resume before any person did. It doesn't count keywords. It checks your resume against up to 50 requirements the employer wrote, gives each one a verdict, and shows the recruiter the exact line it used. InterviewerSignal's comparison of AI screening tools, updated November 9, 2026, describes that setup, and it changes what a good bullet looks like. A vague bullet gets marked "undecided," and on Ashby that counts the same as not meeting the requirement.
How does Ashby's AI screen resumes?
Ashby's AI-Assisted Application Review compares your resume to a list of criteria the recruiter wrote for the job. For each one it returns meets, does not meet, or undecided, along with its reasoning and a citation from your resume. Ashby released it to all customers on September 10, 2024, and it's included in every Ashby plan.
Here's how it works, based on Ashby's documentation and the vendor comparisons:
- Criteria come from the employer. A recruiter writes up to 50 per job. Ashby's docs say one AI credit covers one evaluation with up to 50 criteria in it.
- Each criterion is supposed to be checkable from the resume. Ashby tells employers to write requirements that "can be proven by looking at a resume." One example from its docs: "has led design for consumer saas products with millions of users."
- Every verdict comes with a citation. The recruiter can see which sentence of your resume produced each "meets."
- There's no overall score. Ashby says it "never 'ranks' or gives numerical ratings to applicants" and redacts personal details before the resume reaches the model.
- The recruiter can sort by one column. It's called the AI Job Criteria Met Percentage: the share of criteria a candidate met. Sorting by it moves the closest matches to the top of the queue.
That last point matters most. Ashby doesn't rank you, but the recruiter can sort by that percentage, and the result works like a ranking.
Why does "undecided" hurt as much as "does not meet"?
The percentage counts only criteria marked "meets." An "undecided" adds nothing, exactly like a "does not meet." If the model can't find clear evidence for a requirement, you lose that requirement's share of the score even when you actually have the experience.
This is where well-qualified people lose out. "Undecided" means the model couldn't confirm the requirement from what you wrote. A bullet like "Partnered cross-functionally to drive design outcomes across product surfaces" sounds senior, but it doesn't say whether you led, whether the product was consumer SaaS, or whether it had millions of users. So the verdict is undecided, and you get no credit.
How much does it cost? One screening vendor advises employers to write five to eight specific, resume-verifiable criteria. With a list that short, each criterion carries a lot of weight.
Each requirement is one-eighth to one-fifth of the AI Job Criteria Met Percentage, and undecided scores zero.
| Criteria on the job | Value of each criterion | Score with one "undecided" (all others met) |
|---|---|---|
| 5 | 20 points | 80% |
| 6 | 16.7 points | 83% |
| 8 | 12.5 points | 87.5% |
So a strong candidate with two fuzzy bullets can end up below someone with less experience who wrote plainer ones. The model isn't judging seniority. It's checking whether there's evidence it can cite.
Where do Ashby's criteria come from?
The recruiter writes them, and Ashby can suggest criteria based on the job posting. That makes the requirements list in the posting your best guess at what you're being scored against. Treat each bullet under "Requirements" or "What you'll bring" as a probable criterion.
There's one complication. Employers don't always write good criteria. InterviewerSignal's comparison warns that "vague or inflated must-haves become automated rejections at scale." When a posting asks for "a bias for action" or "startup DNA," the matching criterion is hard to prove from any resume, so many applicants will land on "undecided" for it. You can't fix their wording. You can make sure every requirement that can be proven has a line on your resume that proves it.
In practice:
- Copy the requirement bullets from the posting into a list.
- Next to each one, write the single resume line that proves it.
- If there's no such line, write one. If you honestly don't have the experience, leave it blank and accept the "does not meet."
Doing this for one application is manageable. Doing it for every application in a week is where people give up and send the same resume everywhere. That repeat work is what Refolk handles: paste the posting, and I rewrite your own resume so each requirement has a line of evidence behind it, taken from your actual history and not made up to match.
Why does keyword stuffing backfire on Ashby?
On a reviewer that cites its sources, every "meets" points to a specific line of your resume. If that line is a hidden keyword block or a bare skills list, the recruiter sees it. A trick that might get past a keyword counter becomes visible evidence against you here.
Lots of candidates still try it. Greenhouse's AI-in-hiring research, published October 21, 2026, found:
- 62% of US candidates had tried at least one tactic to get past automated screening.
- 54% used AI to work out what the system was looking for.
- 40% added skills or keywords they don't have.
- About a third of those who tried tactics added hidden text aimed at an AI reviewer. That's roughly 21% of US respondents.
Greenhouse, October 2026. On a citation-based reviewer, the recruiter sees which line the AI matched.
Here's why it backfires. Say you paste a block of keywords in white text at the bottom of the page. The model may mark a criterion as met and cite that hidden line. The recruiter clicks the citation and finds a keyword dump with no job, project, or result attached. The best case is that they ignore it. The worst case is that they stop trusting everything else on the page.
A visible skills list has a milder version of the same problem. As one Ashby resume guide says, a bare skills list can name a tool, but it can't show the AI, or the recruiter reading the citation, that you used it. The skills section can stay. Just don't let it be the only evidence for anything.
Greenhouse CPO Sharawn Tipton described the situation: "For the first time I can remember, neither side thinks hiring is working." A plain sentence of evidence is the one thing that holds up for both sides: the model can cite it and the recruiter can believe it.
A citation-based reviewer turns every keyword trick into a highlighted line on the recruiter's screen.
Where will you run into Ashby?
Fewer recruiters list Ashby than Greenhouse, but the ones who do work at companies like Stripe, Waymo, Drata, and Anchorage Digital. In Refolk's index of professional profiles, 38 US recruiters and talent-acquisition professionals list Ashby as a skill, compared with 184 for Greenhouse.
| Segment (Refolk's index) | Count | Notes |
|---|---|---|
| US recruiters listing Ashby | 38 | Top titles: Recruiter (12), Technical Recruiter (6) |
| US recruiters listing Greenhouse | 184 | Top employer: Meta (3) |
| UK recruiters listing Ashby | 2 | Sunrise Robotics, Reaktor |
| Greenhouse to Ashby ratio, US | 4.8x | Derived |
| Ashby share of Ashby + Greenhouse, US | 17% | Derived (38 / 222) |
A few caveats. These counts only include people with the titles Recruiter, Technical Recruiter, or Talent Acquisition who explicitly list an ATS as a skill, so they're a floor and a relative signal, not the size of the market. The UK figure of 2 is too small to support any conclusion.
The more useful finding is where those 38 work. Ashby-tagged recruiters in the index are at Stripe, Waymo, Drata, Anchorage Digital, brightwheel, and Metropolis Technologies, mostly venture-backed and growth-stage companies. If your target list looks like that, expect some jobs.ashbyhq.com links, and expect your resume to be checked one criterion at a time.
How do you write one line per requirement?
Write one plain sentence for each requirement that names what you did, the context it matches, and a concrete measure. The model needs the requirement's key nouns in a sentence it can cite. The recruiter needs that sentence to be believable when they click through.
The four parts of a citable bullet
- A verb that matches the requirement's level. If the posting says "led," write "led," not "contributed to" or "supported."
- The requirement's context, in its own words. "Consumer SaaS," "B2B payments," "Series B." Use the posting's term when it's accurate.
- A scale or result. Users, revenue, team size, latency, the number of systems. Use a number you'd defend in an interview.
- One requirement per bullet. A bullet covering three criteria can get "undecided" on all three because the model can't tell which part proves which.
Before and after, using Ashby's own example criterion
Criterion: has led design for consumer saas products with millions of users
| Version | Bullet | Likely verdict |
|---|---|---|
| Vague | Drove design outcomes across product surfaces with cross-functional partners | Undecided |
| Keyword-stuffed | Design leadership, consumer SaaS, scale, millions of users | Meets, but the citation is a list |
| Citable | Led product design for [app name], a consumer SaaS product with [N] million monthly users, as sole design lead | Meets, with a clear citation |
The citable version isn't more impressive. It's easier to verify. Every word of the criterion has a matching phrase, and the recruiter who clicks the citation sees a real job with a real number.
Checking that coverage by hand gets tedious after a few applications. Refolk scores your fit before you apply: it marks which posting requirements your resume clearly supports and which are still vague. If a requirement keeps coming back empty, that tells you to skip the job or find the real experience you haven't written down yet.
What doesn't Ashby's AI do?
It doesn't make the hiring decision. Ashby's candidate notice says the AI makes no automated hiring decisions, and a person reviews your application. Most rejections on Ashby come from other features or from a recruiter's bulk action, not from the model.
Things to know before you hit submit:
- Auto-reject is a separate feature. It checks your answers to application-form questions against conditions the employer set, and it runs only when you submit. It's a rule applied to your answers, not a judgement of your resume, so a rejection within seconds most likely means a form answer triggered it.
- You can opt out of AI review. Ashby's notice says opting out "will not negatively impact your application, which will be reviewed manually."
- Bulk actions are the real cutoff. Recruiters can filter by any combination of criteria, sort by the met percentage, and act on the filtered list all at once. Your risk is landing below whatever cutoff a busy person sets on a Tuesday afternoon.
- File size can break parsing. Ashby accepts uploads up to 50 MB, but a resume has to be 16 MB or smaller to be processed for full-text search and to auto-fill application fields. A design portfolio exported as a large PDF can upload without errors and still miss both.
Even if you do all of this, many applications still go unanswered. Greenhouse's October report found that 81% of candidates had been rejected in the past year without ever being contacted. A clean, citable resume moves you up the queue, but a queue sorted by percentage still means many qualified people get archived without a reply.
For the applications you do send, the approach is the same every time. Read the requirements as a criteria list, write one line of evidence for each, and leave out anything you can't stand behind, so that whichever line the model cites holds up when a person reads it.
FAQ
Does Ashby's AI automatically reject resumes?
No. Ashby's candidate notice says the AI makes no automated hiring decisions, and a person reviews the results. Instant rejections usually come from auto-reject rules on your application-form answers, which run at submission and don't look at your resume. Later rejections usually come from a recruiter filtering or sorting by the AI Job Criteria Met Percentage and acting on the list in bulk.
What does "undecided" mean in Ashby AI-assisted application review?
It means the model couldn't confirm from your resume whether you meet the criterion. Because the met percentage counts only "meets" verdicts, an undecided result adds nothing to your score. On a job with five to eight criteria, each undecided costs 12.5 to 20 percentage points.
Should I add hidden keywords for Ashby's AI?
No. Ashby cites the resume text behind every verdict, so if hidden text produces a "meets," the recruiter clicking the citation sees the hidden line. Greenhouse found that roughly a third of the 62% of US candidates who tried screening tactics used hidden text, and on a reviewer that shows its citations, that is the easiest trick for a recruiter to spot.
How do I know which criteria an Ashby job uses?
You can't see the list, but criteria can be suggested by AI from the job posting, and Ashby tells employers to write requirements that can be proven from a resume. Treat the posting's requirement bullets as the criteria list and write one plain sentence of evidence for each one you actually meet.