Should a Candidate Be Flagged for Using Grammarly on a Take-Home Assessment?
TL;DR: No — most hiring teams draw the line between assistive tools (Grammarly, spellcheck, a calculator, an IDE's autocomplete) and generative reasoning tools (ChatGPT solving the actual problem for you), and only the second category gets flagged when undisclosed. The fix isn't banning every tool with "AI" in the name — it's writing an explicit take-home assessment policy that lists what's allowed by category, so candidates and graders aren't guessing.
The claim
Confusion here isn't really about Grammarly — it's that most companies never wrote down where the line sits, so recruiters and hiring managers each draw it differently. That inconsistency creates both false-positive flags (penalizing a candidate for spellcheck) and false-negative gaps (missing a candidate who had ChatGPT write their entire solution) on the exact same assessment.
The evidence
Karat has reported that 80% of candidates use LLMs during code tests even where explicitly banned — a number that only makes sense in a world where policies exist on paper but aren't consistently enforced or clearly scoped. CodeSignal separately found technical assessment cheating roughly doubled year over year (16% to 35%). Neither data point distinguishes "used Grammarly to fix a typo" from "had an LLM write the solution," which is exactly the ambiguity that causes hiring teams to either over-flag minor tool use or under-flag the cheating that actually matters.
Categorizing take-home assessment tools
| Tool category | Example tools | Typical policy | Why |
|---|---|---|---|
| Assistive/editing | Grammarly, spellcheck, IDE autocomplete/syntax highlighting | Allowed by default | Doesn't solve the problem for the candidate — corrects mechanics, not reasoning |
| Reference/lookup | Official docs, Stack Overflow, Google search | Usually allowed, sometimes disclosed | Reflects real on-the-job behavior for most roles |
| Generative reasoning (undisclosed) | ChatGPT, Claude, Copilot writing full solutions | Flagged/disqualifying if the assessment is meant to measure unaided ability | Bypasses the actual skill being tested |
| Generative reasoning (disclosed, open-book AI assessment) | Same tools, used transparently in an "AI-allowed" format | Allowed, scored differently | Measures a different skill — how well a candidate directs AI tools, which is itself valid for some roles |
Step-by-step: writing a take-home tool policy candidates can't misread
- Decide what the assessment is actually measuring. If it's unaided problem-solving ability, generative tools undermine the signal; if it's how well someone works with AI (increasingly relevant for many roles), design an explicit AI-allowed format instead.
- List allowed and disallowed tools by category, not by brand name — "spellcheck and syntax tools: allowed; AI tools that generate code or written answers: not allowed unless this is an AI-allowed assessment" is clearer than trying to enumerate every product.
- State it before the candidate starts, in writing, in the assessment instructions — not buried in a general policy page.
- Match your proctoring configuration to the stated policy. If Grammarly is allowed, your monitoring shouldn't generate noise flags for spellcheck-pattern typing; if it's an AI-allowed assessment, disable generative-tool flags for that req entirely.
- Train graders on the distinction so a candidate who ran spellcheck isn't scored the same as one who submitted an LLM-generated solution.
- Give candidates a way to disclose tool use. A "tools I used" field on the submission form reduces ambiguity and rewards honesty rather than punishing it retroactively.
FAQ
Is there an industry-standard line for what counts as cheating on a take-home? Not a formal standard — it varies by company and even by role — which is exactly why writing your own explicit policy matters more than trying to match an external norm.
Should spellcheck ever be disallowed? Rarely, and only in specific cases like assessing raw writing ability for a copywriting or technical-writing role, where spelling/grammar mechanics are themselves part of what's being evaluated.
What about IDE features like autocomplete that use AI under the hood? Most modern IDEs use some machine learning for basic autocomplete — the meaningful distinction for policy purposes is between minor completion suggestions and a tool that generates entire functions or solutions from a prompt.
How do you handle a candidate who used ChatGPT but didn't realize it violated policy? If the policy wasn't stated clearly, that's a process failure on the company's side — the fairer response is to clarify the policy going forward and give the candidate the benefit of the doubt for that instance, rather than treating an ambiguous policy as grounds for disqualification.
Can an "AI-allowed" assessment format actually be useful for hiring? Yes — for many roles, how effectively a candidate directs, verifies, and iterates with an AI tool is now a real job skill, so some companies run a separate assessment track that explicitly permits and evaluates AI-assisted work rather than banning it outright.
By Pinal Dave Last updated: August 4, 2026