Keep Your AI Consistent with Duplicate FAQ Detection

When you add a Custom FAQ, Rep AI checks your existing Custom FAQs for questions that mean the same thing — and tells you about them before you publish. You see the matching FAQ right inside the Add FAQ window, can read its full answer, and can decide whether to update the FAQ you already have or add a new one.

This keeps your knowledge base clean and your AI's answers consistent. When two FAQs ask the same question but give different answers, your AI has no way to know which one you consider correct — so the same customer question can get different answers at different times. Catching the overlap while you're still typing means you keep one clear source of truth for every topic.

1. Go to AI Training → Knowledge Base

2. Open the Custom FAQs tab

3. Click to add a new FAQ, or open an existing one to edit it

The check runs automatically inside the Add and Edit FAQ windows. There is nothing to switch on and nothing to configure.

As you type in the Question field, Rep AI quietly looks for existing Custom FAQs that ask something similar. The check starts once your question reaches eight characters and waits for a brief pause in your typing, so it never interrupts you mid-thought. While it runs, you'll see "Checking for similar FAQs…" below the Question field.

The check compares meaning, not wording. "What payment methods are accepted?" and "Which payment methods do you accept?" are recognized as the same question even though they share almost no words.

If nothing similar exists, no message appears and you carry on as normal.

A warning appears between the Question and Answer fields. Your draft is untouched and you keep full control of what happens next.

One match — the callout reads "Looks like you already have a similar FAQ" and shows the existing question with the date it was added. You can:

• View existing FAQ — open the full question and answer

• Continue anyway — dismiss the warning and keep writing your new FAQ

Several matches — the callout reads "Found 2 similar FAQs already" (or 3) and lists each matching question with its date. Click any row to open it. If more of your FAQs matched than are listed, a note in the corner tells you how many were found, for example "Showing 3 of 5".

You can also close the warning with the × in its corner at any time.

Opening a match shows you the full existing FAQ — the Existing question, the Existing answer, and when it was added and last modified. From here you have two choices:

• Edit this FAQ instead — closes your new FAQ and opens the existing one for editing, already filled in with its current question and answer. This is usually the right move when the existing FAQ simply needs its answer brought up to date.

• Create new anyway — keeps your draft and dismisses the warning, so you can publish your new FAQ as a separate entry.

When several FAQs matched, a Back to all matches link returns you to the list.

Choosing Edit this FAQ instead replaces the draft you were writing. A short message confirms "Discarded your draft." alongside an Undo button. Click Undo within a few seconds and your original question and answer come straight back.

Rep AI never blocks you from publishing. If you click Publish while a warning is still on screen and you haven't dismissed it, one confirmation step appears:

Publish a possible duplicate?

A similar FAQ already exists. Publishing both may cause the AI to answer inconsistently.

Choose Review existing to go back and take another look, or Publish anyway to add your new FAQ as it stands. Pressing Escape is the same as choosing Review existing — your draft, and the warning, stay exactly as they were.

If you've already dismissed the warning with Continue anyway or Create new anyway, publishing goes straight through with no extra step.

The check works the same way when you edit, with one sensible difference: the FAQ you're editing is never reported as a match against itself. And simply opening an FAQ to change its answer or fix a typo won't trigger anything — the check only runs once you actually start changing the question text.

• Custom FAQs only. The check looks across your Custom FAQs. File Sources, URL Sources, and Promotions aren't included.

• Your FAQs only. Matches always come from your own knowledge base.

• Up to three matches are shown at a time, with the closest match first.

• Existing FAQs are left alone. The check flags overlap at the moment you're creating or editing, so it's worth reviewing your Custom FAQs list directly if you'd like to tidy up FAQs that already cover the same ground.

• It never gets in your way. If the check can't complete, no message appears and publishing works exactly as it always does.

• Treat a match as a prompt to consolidate. One well-maintained FAQ per topic gives your AI a single, unambiguous answer to work from.

• Update rather than duplicate when a policy changes — editing the existing FAQ keeps your answer history in one place.

• Do add a genuinely different FAQ when the questions only look similar. Two questions can be phrased alike but need distinct answers, and Create new anyway exists for exactly that.

• Check your work in Test & Train. After consolidating FAQs on a topic, ask the question there and open the Sources dropdown to confirm your AI is drawing on the FAQ you expect.

I have two similar FAQs but no warning appeared.
The check starts after eight characters and compares meaning rather than wording. If the two questions cover the same topic in genuinely different terms, they may not be close enough to flag. You can always search your Custom FAQs list directly before adding a new entry.

The warning appeared for a question that isn't a duplicate.
The warning is only ever a heads-up — nothing is blocked. Choose Create new anyway or Continue anyway and publish as planned.

My AI is giving different answers to the same question.
This is the usual sign of overlapping FAQs on one topic. Open the Custom FAQs tab, look for entries covering the same ground, keep the one with the correct answer, and remove the rest. Then confirm the result in Test & Train.