Treat it as two separate problems. AI-written applications are a ranking problem: polished writing no longer tells you anything, so test substance with role-specific questions and a structured first interview. Fake applicants are an identity problem: check that contact details and work history hold together, hold at least one live, camera-on conversation, and verify identity and work authorization before day one, as the FBI has advised employers in its 2022 and 2025 warnings.
Small companies used to rely on cheap signals: a cover letter that showed effort, a resume without typos, a candidate who sounded prepared on the phone. In 2026 all three can be produced in seconds. That does not mean most applicants are fake. It means the signals stopped working, and you need checks that do.
It helps to separate two problems that often get mixed up. Most applicants who use AI are real people who want the job and used a tool to write better. A much smaller group are not who they claim to be at all. The first group needs better questions. The second needs verification.
None of this is cheating in itself. Plenty of strong candidates use AI to polish a resume, the same way they once used a friend or a template. What matters is whether they can do the work, and that is something you can test.
Fake applicants are mostly a remote-work problem, and the clearest public guidance comes from the FBI. In June 2022 it warned that people were using deepfakes and stolen personal information to apply for remote IT, programming and database jobs, some with access to customer data and financial systems, and that interviews showed signs of voice spoofing, such as lip movements that did not match the audio (FBI IC3, June 2022).
In January 2025 the FBI updated its warnings about North Korean IT workers who get hired under false identities and later steal or extort company data. Its advice to employers includes verifying identity during interviewing and onboarding, watching for AI and face-swapping tools in video interviews, cross-checking HR systems for applicants who share resumes, phone numbers or email addresses, and doing as much of hiring and onboarding in person as possible (FBI IC3, January 2025).
If you hire for on-site, hourly roles, identity fraud is a smaller risk; you will meet the person on day one. If you hire remote staff with access to systems, money or customer data, treat it as a real risk regardless of company size.
Add one short question to the application that only someone who has done the work can answer well. For a bookkeeper: "Which accounting software have you closed a month in, and what usually breaks at month end?" For a warehouse lead: "How did you handle a shift when two people called out?" A chatbot can produce an answer, but a generic one, and the gap between a generic answer and a lived one is easy to see.
Ask every candidate the same four to six questions about real situations, then follow up: "What would you do differently?" "What did your manager say afterwards?" Follow-ups are where scripted answers run out. An asynchronous first interview by video, phone or chat lets you do this for every applicant who passes the basics, not just the five you have time to call.
If the resume says three years running payroll and the interview answer about payroll is vague, that mismatch is your signal. It is a better one than any judgment about the writing style.
Detectors that estimate whether text was AI-written are useful for deciding where to ask a follow-up question. They are not proof, and they can flag people who simply write in a plain or formal style. Never reject someone on a detector score alone; use it to decide what to probe in the next conversation.
1. Two minutes: scan the application for duplicate contact details or resume text across your other applicants.
2. Three minutes: read the interview answers next to the resume and note any claim the answers don't support.
3. Five minutes: in the live interview, ask one unscripted follow-up about each key claim.
4. Five minutes: call one reference through a number you found yourself.
Apply the same routine to every finalist. That keeps it fair, and it keeps it from depending on anyone's accent, name or appearance, which is both the right thing and the legally safe thing. Federal law prohibits discrimination based on national origin and citizenship status, and singling out some applicants for extra checks is how good intentions turn into a complaint.
HRBlade's AI screens every applicant and runs the first interview by video, phone or chat, then hands you the top five with quotes from their answers. Every score points to what the candidate actually said, so a polished resume with thin interview answers stands out. AI-written answer detection flags responses worth a follow-up question, and opt-in integrity signals add context when you turn them on. None of it decides anything: a person reviews the shortlist and makes the call.
One thing HRBlade deliberately does not do: analyze faces, voices or emotions. It scores the content of answers, not how someone looks or sounds, and that is written down in our no emotion recognition attestation and the trust center. For a small company, that also keeps you away from the kind of AI analysis that draws the most legal attention (see AI interview laws for small businesses).
See how this works for a company with one HR person on the small business page. Sign up free and explore with sample candidates; add a card to go live; the first charge comes after 24 hours, and you can cancel anytime. The Small Business plan is $99 a month, or $79 a month billed annually (see pricing).
No. Detection gives a probability, not proof, and it can flag people who simply write formally. Use it to decide which follow-up questions to ask, never as the reason to reject someone.
Usually not. Many strong candidates use AI to polish their writing. Test whether they can do the work with role-specific questions and a structured interview instead.
There is no reliable public count for small businesses. The FBI has warned since 2022 about applicants using stolen identities and deepfakes, mostly for remote roles with access to systems or data.
A live, camera-on conversation with unscripted follow-up questions before an offer, followed by identity and work authorization verification before any system access. Apply the same checks to every finalist.
No. HRBlade scores the content of answers and does not analyze faces, voices or emotions. It offers AI-written answer detection and opt-in integrity signals, and a person makes every decision.
The information on this website is provided for general information only and does not constitute an offer or a binding commitment. The binding agreement is the Terms of Use at workspace.hrblade.com/terms; where anything on this site is inconsistent with them, those Terms prevail. Product features, availability and figures shown here may change without notice.
