Content Engagement — strategic vision

Date opened: 2026-07-10 Status: Vision + data-gathering only. Not yet a spec. Not yet a build. Why this matters (Arijit's framing, verbatim in spirit): this could define Algolia's next business strategy — moving from a search/discovery platform for eCommerce and Retail/product search into ContentSearch, and from ContentSearch into a machine for Content Engagement.

The core idea

Engagement, not just answers. A visitor asking a question on a website shouldn't just get an answer — the agent should discover intent, understand context, and engage the way a human companion would: proactively, contextually, across a journey.

Flavors of engagement named so far: - Conversational discovery — chat that answers AND discovers (the discovery work already invested in matters here — NLP, grounding, being "fully grounded"). - Journey-embedded engagement — not confined to one chat box. Embedded through different algolia.com journeys: reading a case study, viewing the pricing page, landing on the homepage. Each context should be able to engage differently. - Context/personalization — knowing what page/context the user is in and offering something relevant to that specific moment, not a generic chat widget. - Conversion-oriented — the point of engagement is to bring the visitor into the fold: book a call, start building, ask something deeper. Engagement is instrumental to a business outcome, not just a UX nicety.

Backend = Algolia. Delivery layer = Agent Studio agents (Algolia's own product).

POC scope (as currently framed)

• Journeys targeted (parallel, not sequential): case study pages, pricing page, homepage/landing page.
• Deployment for the POC: standalone demo environment that simulates those page types. Not injected into live algolia.com — lower risk, faster iteration, no production exposure while the concept is unproven.
• Audience / success bar: broad market / product-market validation. Not scoped to convincing one exec or landing one named customer — the bar is "does this concept hold up broadly."

Explicitly open — not yet decided

Relationship to the existing AC2 build. Algolia-Central2 already has a live, tested 3-agent Agent Studio architecture (General / Developer / Marketer-Merchandiser) on app 0EXRPAXB56, built and verified as of session 7 (2026-07-01) — see SESSION.md. That build already answers grounded questions from algolia.com content via source-scoped retrieval on one neural index. Arijit's explicit answer when asked whether Content Engagement extends that build or runs parallel to it: "not sure yet — that's part of what we're figuring out." Do not force a merge decision. Capture both threads; let the data-gathering (below) inform which way this goes.

Data-gathering thread opened in parallel (2026-07-10)

Distinct from the vision itself — this is grounding work to inform where/how engagement should actually show up.

1. Analytics API pull — full breadth and depth of real queries against ALGOLIA_WWW_PROD_V2 (what are people actually asking/searching for on algolia.com today).
2. Insights API pull — click-through rate, conversion rate, no-results rate for the same index — behavioral signal, not just query text.
3. Index x-ray — study ALGOLIA_WWW_PROD_V2's structure end-to-end: what fields exist, how rich vs. thin each is, what's missing, what enrichment would improve quality. Precedent: AC2's own index (AC2_WWW_MULTI_NEURAL) went through a body-enrichment A/B in session 5-6 — verdict was no answer-quality lift from making body searchable (see feedback-rounds1-ab-double-noise / session 6 in SESSION.md). Any enrichment argument for ALGOLIA_WWW_PROD_V2 should reckon with that prior null result rather than assume enrichment automatically helps.
4. Screenshot sweep — full-page screenshots of representative algolia.com pages (case study, pricing, homepage) to ground a joint UI/UX brainstorm: what should pop up, what should the conversation be, at what point in the journey.
5. GA behavioral data — Looker Studio dashboard Arijit has access to (datastudio.google.com/u/0/reporting/b05b44b9-43bd-436b-8dd3-c92729a93a93/page/Cr7KB). Confirmed NOT reachable via plain WebFetch — redirects straight to Google login (private report). Two live options: (a) drive it via an authenticated Chrome browser session if Arijit's already logged in there, with the caveat that concurrent Chrome DevTools MCP sessions can lock each other out of the same browser profile; or (b) Arijit exports CSV/Sheets/PDF directly, which is more reliable and gives structured data instead of a dashboard screenshot to interpret.

Grounding note (resolved without asking Arijit)

ALGOLIA_WWW_PROD_V2 is not part of the AC2 build app. It lives in the VISIBILITY app, 1QDAWL72TQ — documented in RAG/Algolia-Central2/CLAUDE.md as "Algolia's live keyword app (incumbent), READ-ONLY — benchmark target — never write." This is the actual live production search index behind algolia.com today (the thing AC2 is trying to beat). Resolved by reading the project's own CLAUDE.md instead of asking — the app ID and read-only status were already documented.

Next

Run the Analytics/Insights/index-x-ray pulls against 1QDAWL72TQ / ALGOLIA_WWW_PROD_V2 (read-only, safe). Synthesize data-shape findings. Then a joint brainstorm — screenshots + query data + GA data together — on where in the journey engagement should actually surface, and what it should say/do at each point.