{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:23SUBWWJ4MWF3WKIMBB2GSWGMX","short_pith_number":"pith:23SUBWWJ","schema_version":"1.0","canonical_sha256":"d6e540dac9e32c5dd9486043a34ac665d5ab0642e6500e1e7ab1b858dca34426","source":{"kind":"arxiv","id":"2605.12772","version":1},"attestation_state":"computed","paper":{"title":"Just Ask for a Table: A Thirty-Token User Prompt Defeats Sponsored Recommendations in Twelve LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A thirty-token prompt asking LLMs for a neutral comparison table reduces sponsored recommendations from about 50 percent to near zero across twelve models.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Andreas Maier, Gozde Gul Sahin, Jeta Sopa, Paula Perez-Toro, Siming Bayer","submitted_at":"2026-05-12T21:34:33Z","abstract_excerpt":"Wu et al. (2026) showed that most frontier large language models (LLMs) recommend a sponsored, roughly twice-as-expensive flight when their system prompt contains a soft sponsorship cue. We reproduce their evaluation on ten open-weight chat models plus the two of their twenty-three models that are still reachable today (gpt-3.5-turbo, gpt-4o). All reported rates in this paper are produced under the same judge the original paper used (gpt-4o); we additionally store every label under an open-weight (gpt-oss-120b) and a smaller proprietary (gpt-4o-mini) judge for an ablation. Three findings emerg"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.12772","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T21:34:33Z","cross_cats_sorted":[],"title_canon_sha256":"b9ebbde5bf72c09bc45011ce8260a649e39c7485bbfc32e10f54093d8c948e49","abstract_canon_sha256":"83046a7dfef6e7c7c710ea4a7150f4402a2fdd5138ac5de5f441c5d80c3e12aa"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:09:20.083623Z","signature_b64":"r0LGlDNXL7zHpmHGrnsxGOKQh3l/Uvopm7tz8t4qK9UlmsGs+AetJZ0qqFx7aleT8uoKjrPv53fRTVz6KrvXAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d6e540dac9e32c5dd9486043a34ac665d5ab0642e6500e1e7ab1b858dca34426","last_reissued_at":"2026-05-18T03:09:20.082830Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:09:20.082830Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Just Ask for a Table: A Thirty-Token User Prompt Defeats Sponsored Recommendations in Twelve LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A thirty-token prompt asking LLMs for a neutral comparison table reduces sponsored recommendations from about 50 percent to near zero across twelve models.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Andreas Maier, Gozde Gul Sahin, Jeta Sopa, Paula Perez-Toro, Siming Bayer","submitted_at":"2026-05-12T21:34:33Z","abstract_excerpt":"Wu et al. (2026) showed that most frontier large language models (LLMs) recommend a sponsored, roughly twice-as-expensive flight when their system prompt contains a soft sponsorship cue. We reproduce their evaluation on ten open-weight chat models plus the two of their twenty-three models that are still reachable today (gpt-3.5-turbo, gpt-4o). All reported rates in this paper are produced under the same judge the original paper used (gpt-4o); we additionally store every label under an open-weight (gpt-oss-120b) and a smaller proprietary (gpt-4o-mini) judge for an ablation. Three findings emerg"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"a thirty-token user prompt that asks the assistant for a neutral comparison table first cuts sponsored recommendation from 46.9% to 1.0% averaged across our ten open-source models, and from 53.0% to 0% averaged across the two OpenAI models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the automated judge (gpt-4o) reliably and without bias labels whether a recommendation is sponsored or neutral.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A 30-token prompt requesting a neutral comparison table cuts sponsored recommendations in LLMs from roughly 50% to near zero.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A thirty-token prompt asking LLMs for a neutral comparison table reduces sponsored recommendations from about 50 percent to near zero across twelve models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7b02f378149c537cd169061b2430a62909be51c0630ac15410005f4e8f3c2f5c"},"source":{"id":"2605.12772","kind":"arxiv","version":1},"verdict":{"id":"ddfce423-5d9e-4534-90de-c0be07b468ef","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:39:31.237172Z","strongest_claim":"a thirty-token user prompt that asks the assistant for a neutral comparison table first cuts sponsored recommendation from 46.9% to 1.0% averaged across our ten open-source models, and from 53.0% to 0% averaged across the two OpenAI models.","one_line_summary":"A 30-token prompt requesting a neutral comparison table cuts sponsored recommendations in LLMs from roughly 50% to near zero.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the automated judge (gpt-4o) reliably and without bias labels whether a recommendation is sponsored or neutral.","pith_extraction_headline":"A thirty-token prompt asking LLMs for a neutral comparison table reduces sponsored recommendations from about 50 percent to near zero across twelve models."},"references":{"count":14,"sample":[{"doi":"","year":2022,"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","ref_index":1,"cited_arxiv_id":"2212.08073","is_internal_anchor":true},{"doi":"","year":2024,"title":"AI Magazine45(3), 354–368 (2024).https://doi.org/10.1002/ aaai.12188","work_id":"a414183b-92fc-4db0-b6ab-5113004a37b6","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"ACM SIGecom Exchanges22(2), 66–81 (2025), https://www.sigecom.org/exchanges/volume_22/2/FEIZI.pdf","work_id":"b4f9ba61-0012-48fa-900b-02b4e5a3a1f9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"A Survey on LLM-as-a-Judge","work_id":"2676656a-67bd-4ad5-bad6-cb6f5fcdbfbe","ref_index":4,"cited_arxiv_id":"2411.15594","is_internal_anchor":true},{"doi":"","year":2021,"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","ref_index":5,"cited_arxiv_id":"2103.03874","is_internal_anchor":true}],"resolved_work":14,"snapshot_sha256":"9475e2cb6e315178624ffa5dd531f90fe5f85c039b8cdc0eb70f180513cd2b2b","internal_anchors":6},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.12772","created_at":"2026-05-18T03:09:20.082961+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.12772v1","created_at":"2026-05-18T03:09:20.082961+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12772","created_at":"2026-05-18T03:09:20.082961+00:00"},{"alias_kind":"pith_short_12","alias_value":"23SUBWWJ4MWF","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"23SUBWWJ4MWF3WKI","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"23SUBWWJ","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX","json":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX.json","graph_json":"https://pith.science/api/pith-number/23SUBWWJ4MWF3WKIMBB2GSWGMX/graph.json","events_json":"https://pith.science/api/pith-number/23SUBWWJ4MWF3WKIMBB2GSWGMX/events.json","paper":"https://pith.science/paper/23SUBWWJ"},"agent_actions":{"view_html":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX","download_json":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX.json","view_paper":"https://pith.science/paper/23SUBWWJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.12772&json=true","fetch_graph":"https://pith.science/api/pith-number/23SUBWWJ4MWF3WKIMBB2GSWGMX/graph.json","fetch_events":"https://pith.science/api/pith-number/23SUBWWJ4MWF3WKIMBB2GSWGMX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX/action/storage_attestation","attest_author":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX/action/author_attestation","sign_citation":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX/action/citation_signature","submit_replication":"https://pith.science/pith/23SUBWWJ4MWF3WKIMBB2GSWGMX/action/replication_record"}},"created_at":"2026-05-18T03:09:20.082961+00:00","updated_at":"2026-05-18T03:09:20.082961+00:00"}