168. AI Detection and Authenticity
By the end you'll be able to
- Explain how current AI detectors work and why they produce false positives and false negatives.
- Identify prose patterns that increase detection risk (low burstiness, hedged voice, recycled phrasing).
- Use AI for structure and refinement without producing the flat output detectors and reviewers flag.
- Treat honest disclosure as a defensive tool when detection is in play.
AI detection tools are imperfect, inconsistent, and increasingly common in funder review workflows. That combination is a problem. Detectors produce false positives on plain, careful prose. They produce false negatives on lightly edited model output. Funders use them anyway, sometimes as a screening filter and sometimes as a flag for closer human review. Pretending the technology does not exist will not protect your proposals. Understanding it will.
In this lesson you will learn how current detection tools actually work, what signals they look for (perplexity, burstiness, vocabulary patterns), and why those signals are unreliable. You will see why a proposal that reads as monotone, hedged, and generic is more likely to trigger detection than one that carries a real voice, even if both used AI in the drafting process. You will practice techniques that preserve authenticity: writing the spine yourself, using AI for restructuring rather than wholesale generation, and editing aggressively for specificity, anecdote, and idiom. You will also learn when disclosure is your strongest defense, because a flagged proposal is much easier to defend when you have already told the funder how AI was used.
By the end you should be able to use AI throughout your workflow without producing the flat, hedged, recycled prose that gets proposals flagged. Authenticity is no longer a vibe. It is a measurable risk factor, and you can manage it.
Common mistakes
These are the traps learners hit most often on this topic. Knowing them in advance is half the fix.
Asking AI to "humanize" the text as the last step.
Round-tripping through a humanizer often produces prose that still scans as model output and now reads as awkward. Authentic voice has to be built in from the spine, not painted on.
Burying disclosure in fine print.
A disclosure that no human reviewer will read does not protect you. Place it where the funder requires, in plain language, with specifics about how AI was used.
Practice problems
Try each on paper first. Click Show solution only after you've made a real attempt.
- Problem 1Rewrite the following hedged sentence to reduce detection risk and improve voice. Original: "This project will potentially address various challenges that may impact underserved communities by leveraging multiple evidence-based approaches."
Show solution
This project tackles three problems that block first-generation students in West Baltimore from finishing the FAFSA: confusing language, missing tax documents, and no quiet place to fill it out. We address each with a separate intervention, not a single magic program.
Practice quiz
- Question 1Which of the following is most likely to produce a false positive on an AI detector?
- Question 2When is disclosure most useful as a defensive tool?
Lesson 168 recap
AI detectors are imperfect but real. Authentic voice and honest disclosure are the two strongest defenses, and both have to be designed into the workflow.
Coming next: Lesson 169 — AI For Accessibility
Next, we flip the lens and look at how AI can expand access and equity for the communities your proposals describe.
Saved in your browser only — no account, no server.