Teach it your business
What the Review tab shows you
The Review tab is where you read back what your bot took out of your own pages and files.
Before you start
- Your bot has finished learning at least one page, file or FAQ How to check
About 6 minutes. You’ll do three things:
- read what was taken in
- judge the drafted answers
- fix what is wrong
The Review tab is a read-back of your own content. Every time your bot learns a page or a file, Adoomi also pulls the facts out into sections, drafts a few questions those facts would answer, and lists any safety wording it found. The tab shows you all three.
You will find it on Train your bot, on the tab marked Review. Some people call it the review queue. On screen it is just Review.
What lands there
| Block | What it holds |
|---|---|
| Shadow index | A build status line. On live accounts it reads No shadow index version yet. and that is normal. |
| Knowledge bundle | The facts taken from your sources, grouped under Core business info, Services, Pricing and payments, Customer journey, FAQs, Safety and boundaries and Live actions. Items that came from a web page link back to it. |
| Generated Q&A | Questions Adoomi drafted from your content, each with an answer you can edit and three buttons. Highest priority first. |
| Safety rules | Wording about limits and risk that was found in your content, grouped by how serious it is. |
Take a plumber whose site has 30-odd pages. After the first crawl the tab shows a Knowledge bundle with 6 items under Services and 2 under Pricing and payments, and 14 cards under Generated Q&A. Nobody asked those 14 questions. Adoomi wrote them from the pages, so the first job is reading them against what is actually true today.
Nothing here was written by a visitor. Every question on the tab was drafted from the shape of your own content, so a question can look sensible and still be one nobody has ever asked you.
What reviewing actually does
Editing an answer and clicking Approve saves your wording and records who approved it and when. Needs review parks it. Reject bins it. All three save whatever is in the answer box at the time, so a tidy-up before you click is never wasted.
How this fits with gaps and FAQs
| Tab | What it shows | Does it change a live answer? |
|---|---|---|
| Review | What your bot took out of your own pages and files. | No. Your decisions are kept for the next index build. |
| Knowledge gaps | Questions visitors asked that got your fallback message. | No. It tells you what is worth writing. |
| FAQs and Custom replies | Answers you typed yourself. | Yes, as soon as the new row reads ready. |
Read together they make one loop. Review tells you what your bot understood. Knowledge gaps tells you where that was not enough. An FAQ or a custom reply is how you close the difference.
| Option | Best for |
|---|---|
| Approve | The answer is right, or right after your edit. The line under the question then reads approved. |
| Needs review | You want to come back to it. The line reads needs review. |
| Reject | Nobody would ask this, or the answer is wrong. The line reads rejected. |
Work through the tab
In the bot sidebar, click Train your bot, then the Review tab.
Read the Knowledge bundle sections against what is true today.
The number beside each section heading is how many items it holds. Items taken from a web page also show a Source link back to it.
Scroll to Generated Q&A and start at the top card.
The line under each question shows its current status and a priority number. The list is ordered by that number.
Fix the wording in the answer box where it is close but not quite right.
Click the button that matches your decision on that card.
All three save whatever is in the answer box at the time.
Work down the list, noting anything a customer would actually ask.
Those are the ones worth writing out properly on the Custom replies tab, where they will reach a visitor.
A wrong fact in Knowledge bundle is a wrong fact on the page it came from. Fix the page, add or re-add it as a source, and the block rebuilds from the corrected version.
If it didn’t work
The tab says No structured review assets have been generated yet.
- Check
- That shows on a bot that has not finished learning anything, or one whose sources all went in before this screen existed.
- Fix
- Open the Retrain tab and click Update bot with latest. Come back to Review once the run finishes.
- Resolved when
- Sections appear under Knowledge bundle with counts beside them.
Shadow index says No shadow index version yet.
- Check
- That card reports an internal build that is switched off on live accounts. It is not a fault and it is not waiting on anything you do.
- Fix
- Leave it. Your bot's answers come from your pages, documents, FAQs and custom replies, none of which go through that build.
- Resolved when
- You can carry on with the rest of the tab, which works whether or not that card ever fills.
I approved an answer and my bot still says something else
- Check
- Ask the same question in the widget and compare. Approving records your decision for the next index build. It does not put your wording in front of a visitor today.
- Fix
- Copy the answer you approved into the Custom replies tab and click Add reply. That one does reach visitors, as soon as the row reads ready.
- Resolved when
- The bot replies with your wording.
Update failed. appears under a card
- Check
- That is the save on that one card failing, most often a dropped connection.
- Fix
- Reload the page and click the button again. If it keeps failing, email hello@adoomi.ai with the bot's name and the question on the card.
- Resolved when
- The line under the question shows the status you picked.
The tab says Structured review data could not be loaded.
- Check
- The page opened but the review data behind it did not. The rest of Train your bot still works normally.
- Fix
- Reload the page. If the banner is still there after two or three tries, email hello@adoomi.ai with the bot's name.
- Resolved when
- The banner is gone and the sections list again.
A question on the tab is one nobody has ever asked me
- Check
- These questions were drafted from your content, not from real chats. A page about parking produces a parking question whether or not a visitor cared.
- Fix
- Click Reject on it and spend the time on the Knowledge gaps tab instead, which only lists questions visitors really did ask.
- Resolved when
- The card is rejected, and your writing time goes on questions with an Asked count behind them.
What this can’t do
- Approving an answer does not change what your bot says to a visitor. Write it as a custom reply to do that.
- You cannot add a card yourself. Everything on the tab was generated from sources you added.
- A Knowledge bundle item cannot be edited. Fix the page or file it came from, then retrain.
- Shadow index stays empty on live accounts, so its counts never fill in.
- The questions here come from the shape of your content, not from real visitors. Knowledge gaps is the tab that reflects what people actually asked.