Turn Algolia zero-result searches into Intercom Help Center drafts

By General Input

Every Monday morning, last week's dead-end searches on your site become draft help articles sitting in Intercom, ready for a human to review and publish.

Integrations

  • Algolia
  • Intercom

Type

Agentic Task

Categories

  • Customer Support
  • Marketing

Every Monday at 9am, turn last week's Algolia zero-result searches directly into draft Intercom Help Center articles so support content actually closes the gaps customers keep hitting.

Trigger: cron, every Monday at 09:00 in the workspace timezone.

Step 1. Call Algolia's Analytics: Top Searches with No Results for the trailing 7 days on the configured index. Pull the search term and the search volume (count) for each entry. Keep only terms whose volume is at or above a configurable threshold (default 5 searches in the week).

Step 2. Skip anything that looks like personal information (email addresses, phone numbers, order numbers, names that read like a customer identity) or an obvious misfire (single characters, random keyboard mashing, gibberish, adult or abusive language). Do not draft an article for these.

Step 3. Cluster near-duplicate queries into a single group before drafting. Typos, plurals, word order swaps, and small phrasings of the same underlying intent become one cluster (for example "refund policy", "refund policys", "how to refund", and "can i get a refund" are one cluster). Pick the clearest phrasing as the cluster label. One draft per cluster, never one article per typo.

Step 4. Call Intercom's List All Articles and page through the full library. For each cluster, compare against existing article titles and summaries and skip the cluster if it is already covered by a live or draft article. Be conservative: if you are unsure whether it is truly covered, keep the cluster and draft the new article anyway.

Step 5. For each genuinely new cluster, draft a full help article with these fields: a clear customer-facing title, a short intent summary (one to two sentences on what the customer is trying to do and what the article answers), three to five subheadings that structure the answer, placeholder body copy under each subheading that sketches what the human author should fill in, and a small set of suggested tags derived from the cluster keywords.

Step 6. For each drafted article, call Intercom's Create an Article with state set to draft. Never publish. Include the title, the intent summary as the description, and the subheadings plus placeholder body copy as the article body. Attach the suggested tags. Draft-only publishing is a hard rule so a human always reviews before anything ships to customers.

Configurable knobs: the Algolia index to analyze, the minimum weekly search-volume threshold, the day and time the workflow runs, an optional Intercom collection or author to assign drafts to, and any custom skip list of query patterns (for example the company's own brand terms if you do not want articles drafted for them).

Output: a short summary at the end of the run listing each new draft created (title and cluster of queries it covers), each cluster skipped as already covered, and each cluster skipped as personal information or a misfire, so the support lead can spot-check the run.

Related prompts

Explore more prompts
Rank your Intercom queue by what each customer pays youCatch the trials quietly dying before the card gets chargedAudit what Intercom's Fin AI actually resolved before you payTurn shipped Linear work into Intercom product announcementsAnswer Intercom chats in any language, with a meaning checkSee which help center articles have audio in every languageCurate Algolia search results with live Shopify stockTriage and fix Algolia searches that return no resultsCatch Algolia index drift before a campaign goes liveWeekly voice-of-customer report from surveys and support