What changed,
in public.
No customer logos yet — so the proof is the work itself. Every material product and measurement change is logged here, the engine models below are read live from the adapter configuration, and the studies the numbers come from are linked and dated.
Engine model log
Answer-engine sweeps record the exact modelVersion each engine returns on every run. When an engine's version changes between two runs of the same prompt set, the model-version radar compares presence and sentiment across the transition and flags statistically significant shifts — correlation, dated and labelled, never causation. These are the models the adapters currently target:
| Engine adapter | Current model target | Interface |
|---|---|---|
| ChatGPT (API) | gpt-5.6-terra | Provider API, per-call pricing |
| Claude (API, web search) | claude-sonnet-4-20250514 | Provider API, per-call pricing |
| Google Gemini (API, grounded) | gemini-2.5-flash | Provider API, per-call pricing |
| Perplexity (Sonar API) | sonar | Provider API, per-call pricing |
| Manual observation (any surface) | manual-observation | Human-observed answers, pasted from any surface |
Targets are configuration, pinned per deployment; the versions recorded on archived runs are observed values returned by the engines themselves. When a target changes, the transition appears here and in every workspace running sweeps against that engine.
Proof you can check
- Measured AI crawler study — fifteen AI user-agents resolved against the live robots.txt of 87 well-known websites, dated, with the deciding rule behind every verdict.
- The full sample report — a complete worked Professional report: five tabs of query, page, model and plan detail, with illustrative data labelled as such.
- The method — the seven-step evidence process, including what IndexHalo measures directly, what it cross-references, and what it only estimates.
Product changelog
| Date | Change | What it means |
|---|---|---|
2026-08-12 | Design review implemented end to end | The landing page split into focused URLs (/method.html, /sample-report.html), the signed-in workspace gained real routes (/app, /app/reports/:id, /app/sites/:id, /app/studio), the dashboard went delta-first with an activation checklist and actionable failure rows, and reports now lead with the fix, keep the score on screen while scrolling, and print self-explanatory. |
2026-08-12 | Shared design tokens and view primitives | One token file now feeds the marketing pages, the workspace, and the editorial pages, so a colour or type change cannot drift between copies. |
2026-08-12 | Agency deliverable and plan lifecycle | White-label report exports for Agency accounts, plan gating that names the limit before you hit it, and failed-payment recovery. |
2026-08-11 | 87 measured AI crawler reports published | Fifteen AI user-agents resolved against the live robots.txt of 87 well-known websites — every verdict dated and traceable to the deciding rule. Public at /ai-crawler-report/. |
2026-08-11 | Sourced competitor comparisons | Three head-to-head pages that state where the competitor is the better choice and link every claim to the vendor's own pricing page. |
2026-08-10 | Citation Studio | Block-based composer for citation-ready pages: fourteen measurement-backed block types, live scoring as you type, and export to semantic HTML, JSON-LD, llms.txt and robots stance. |
2026-08-10 | Answer-engine intelligence panel | Persona stress tests, defensive probes, citation poaching and parentage, cross-language drift, fair-use audit, and the model-version regression radar across four measurement phases. |
Measurement-affecting changes — scoring weights, signal definitions, evidence classes — are called out in the entry itself, because a score that moved because the ruler changed says nothing about your content.