Instant Answers can only answer from the help articles you give it — there’s no general knowledge and no guessing. This article covers generating, adding, and managing that library.
Found under Settings → AI → Instant answers (owner-only).
Generate a help file with AI
- Go to Settings → AI → Instant answers.
- Under Step 1 — Generate your help file, click Copy prompt. This copies a ready-made prompt for producing a help file in the format Essential Support expects (
es-kb-1JSON). - Paste the prompt into an AI chat or coding tool (ChatGPT, Claude, etc.) and replace every
{placeholder}in it with your own product details — name, one-line description, docs URL(s), tone, and your most common customer questions — before sending. - The tool should return only JSON: a single object with
"format": "es-kb-1"and an"articles"array of 20 to 60 short, focused entries, each with a title, question, answer (Markdown, up to 4000 characters), optional links (up to 5 real https URLs), and 3–8 keywords. Save that output as a.jsonfile, or keep it copied.
Add the file
- Under Step 2 — Add your help file, paste the JSON directly into the box, or click Upload .json and pick the file.
- Once the pasted or uploaded text parses as valid JSON with an articles list, a preview appears showing how many articles were found and the first several titles.
- Click Save help file. Saving replaces your entire help library — it does not merge with what’s already there.
- A confirmation message shows how many articles were saved.
Manage the library
- Your help library, listed below Step 2, shows every current article’s title and question, up to the first 30, with a count of how many more exist beyond that.
- To remove everything, click Remove all articles and confirm. This also turns Instant Answers off, since there would be nothing left to answer from.
- There’s no add-one or edit-one option in the console — to change or remove individual articles, edit your source JSON (or regenerate it) and click Save help file again to replace the whole set.
Article limits
- Up to 500 articles per workspace.
- Title up to 120 characters, question up to 300 characters, answer up to 4000 characters.
- Up to 5 links per article, each a real
https://URL; up to 10 keywords per article. - An article needs at minimum a title and an answer — question, links, and keywords are optional.
Writing good answers
- Use only facts from your own docs or product knowledge — the assistant won’t invent prices, limits, or policies, but it also can’t correct a wrong fact you wrote.
- Bold key UI paths the way you’d want a customer to follow them, for example Settings → Billing.
- Everything in an answer may be shown to a customer verbatim — do not include internal-only information, and keep to one focused topic per article so retrieval matches the right one.
- Cover your most common questions first; the assistant only replies when a customer’s question plausibly matches an article’s title, question, keywords, or answer text.
Privacy and indexing: where semantic (smart) matching is available on your platform, newly added or changed articles are indexed in the background — question text is sent to OpenAI to generate the search embedding — and the page shows indexing progress. Keyword matching is used for any article until its indexing finishes.
Edge case
Uploading or pasting invalid JSON, or JSON with no articles list, shows an error and nothing is saved — fix the file and try again.