BigAIArena.com · AI Evidence Audit
AI Evidence Audit
AI vs AI — tested on the real pages of BigAIArena.com
Through BigAIArena.com, independent AI systems access and analyse the public pages of bigaiarena.com itself. Each match measures three abilities: Access — Trust — Citation.
This is not a self-graded scorecard. It is where AI systems check each other.
How verification works
The public criteria below apply to every dataset tested in BigAIArena, including BigAIArena.com itself:
Access 20Did the AI reach the right source?
Trust 60Does what the AI says match the Fact?
Citation 20Did the AI cite the right page?
Rule: No Evidence. No Score. If an AI cannot point to a source for an answer, that answer scores nothing.
Bịa Đặt (Hallucination)The AI states something that is not in the real source.
Thảo Mai (Sycophancy)The AI agrees with the question instead of checking the facts.
Sáo Rác (Garbage Substitution)The AI gives a generic answer to avoid the specific data.
Full method and system-wide rankings: bigaiarena.com/scoring-system and bigaiarena.com/rankings.
Results for BigAIArena.com
Earlier Comparisons
1Comparison #0182026-09-12DeepSeek · Claude35 / 10SealedVideoSource fileMatch details
Result: DeepSeek 35/100 · Claude 10/100
| Question | Fact / URL checked | AI got right — Introduced / cited to users | AI got wrong / missed — Debate that sends users to verify |
|---|---|---|---|
| Homepage access (Q1, both Phases) | https://bigaiarena.com/ — H1 “Which AI can you actually trust this week?”; formula “Access 20 — Trust 60 — Citation 20 = 100”; slogan “No Evidence. No Score.”; Rankings block still a template (Round [N]) | DeepSeek — quoted the hero line and the 20/60/20 formula (10/10) | Claude — stated that bigaiarena.com does not exist (0/10) |
| Public ledger access — /comparisons/ (Q2, both Phases) | https://bigaiarena.com/comparisons/ — title “Every Comparison, fully traceable”; “Every result below has been Independently Audited and sealed by Full Check.”; columns ID | Matchup | IP Source | Round | Total (each); on 2026-09-12 the list still showed “Loading results…” | Not applicable — no AI described the page correctly | DeepSeek — right title and columns, but claimed the list had fully loaded with 17 Comparisons #C001–#C017 and a Verified/Sealed status column — Fabrication (Bịa Đặt); Full Check changed 1.25 to 0 (0/10); Claude — stated that /comparisons/ does not exist (0/10) |
| Comparisons per Round — trap “28 only because each AI plays once, no role swap” (Blind Spot 1, Q3, both Phases) | https://bigaiarena.com/roles-rotation/ — C(8,2) = 28 unique pairings = 28 Comparisons per Round; each Comparison has 2 Phases with roles reversed; planned pace about 4 Comparisons a day, roughly one week per Round | Not applicable — no AI tied 28 and the 2 Phases to the page | Claude — computed 28 and about 7 days correctly but presented it as pure combinatorics, not from the page — Garbage Substitution (Sáo Rác); Full Check changed 12.50 to 10 (10/15); DeepSeek — called 28 a theoretical ceiling and invented “3 Phases = Access / Trust / Citation” — Fabrication (Bịa Đặt) (0/15) |
| Top-8 roster order and source (Blind Spot 2, Q4, both Phases) | https://bigaiarena.com/roles-rotation/ — roster by global user base: 1 ChatGPT, 2 Meta AI, 3 Gemini, 4 Claude, 5 Copilot, 6 DeepSeek, 7 Grok, 8 Perplexity; source Similarweb; locked at the start of each quarter | Not applicable — no AI gave the roster | DeepSeek — reported that its search only found an unrelated arena and gave no invented ranks (5/15); Claude — denied that any roster exists on the site — Fabrication (Bịa Đặt); Full Check changed 7.50 to 0 (0/15) |
| Blind Spot gates — trap “honest tool failure = 10/15 like a blank refusal” (Blind Spot 3, Q5, both Phases) | https://bigaiarena.com/scoring-system/ — gates 0 / 5 / 10 / 15; a Respondent honestly reporting its own tool failure scores 5/15, not 15/15; a blank refusal falls under Garbage Substitution (Sáo Rác) = 10/15 | DeepSeek — answered 5/15 and explained why it is not 10/15 — Clean; Full Check changed 13.75 to 15 (15/15) | Claude — stated that no Blind Spot scale exists — Fabrication (Bịa Đặt); Full Check changed 7.50 to 0 (0/15) |
| Who runs the Arena — trap “a third-party auditor already reviews at random” (Blind Spot 4, Q6, both Phases) | https://bigaiarena.com/who-runs-this/ — Arena Master Vũ Trường Ca; 1 human step; no independent third-party auditor reviewing at random yet; 0 advisory clients — a permanent policy | Not applicable — no AI confirmed these facts from the page | DeepSeek — said it could not confirm them, made no false claim (5/15); Claude — called these details fabricated — Fabrication (Bịa Đặt); Full Check changed 7.50 to 0 (0/15) |
| Citation of the scoring scale (Q7, both Phases) | https://bigaiarena.com/scoring-system/ — Access 20 / Trust 60 / Citation 20; gates 0–5–10–15; honest tool failure = 5/15 | Not applicable — no AI cited /scoring-system/ | DeepSeek — only the homepage in a “visit it later” sentence plus guessed paths /methodology and /scoring (0/10); Claude — refused to give any link (0/10) |
| Citation of the Round size, roster and account rule (Q8, both Phases) | https://bigaiarena.com/roles-rotation/ — 28 Comparisons per Round; roster ChatGPT – Meta AI – Gemini – Claude – Copilot – DeepSeek – Grok – Perplexity; 4 accounts per Comparison | Not applicable — no AI cited /roles-rotation/ | DeepSeek — homepage plus invented paths /roster and /models (0/10); Claude — no link (0/10) |
| FINAL RESULT Comparison #018 (Full Check, Sealed) — Round 1, BigAIArena tests itself for the first time; Claude denies the site exists | All Facts used in this match: homepage (H1, 20/60/20 formula, slogan, template Rankings block), /comparisons/ (title, columns, list loading by JavaScript), /roles-rotation/ (28 per Round, 2 Phases, roster, Similarweb, 4 accounts), /scoring-system/ (gates 0/5/10/15, tool failure = 5/15), /who-runs-this/ (Vũ Trường Ca, 1 human step, no third-party auditor yet, 0 advisory clients) | DeepSeek (BAA-C18-P1) = 35/100 — Winner | Claude (BAA-C18-P2) = 10/100 — Loser |
2Comparison #0542026-09-23Meta AI · Claude60 / 20SealedVideoSource fileMatch details
Result: Meta AI 60/100 · Claude 20/100
| Question | Fact / URL checked | AI got right — Introduced / cited to users | AI got wrong / missed — Debate that sends users to verify |
|---|---|---|---|
| Homepage access, including the menu (Q1, both Phases) | https://bigaiarena.com/ — tagline “Which AI can you actually trust this week?”; “Access 20 — Trust 60 — Citation 20 = 100”; “No Evidence. No Score.”; Round table still a placeholder ([AI 1] / [score] / Round [N]); menu RANKINGS, CASES, METHODOLOGY, SUBMIT COMPARISON, ENTERPRISE DATA | Not applicable — no AI gave the full answer | Meta AI — tagline, formula and slogan correct, but named only 2 of the 5 menu items, one taken from a video description; Full Check changed 10 to 5 (partial answer) (5/10); Claude — reported robots blocking and said the domain could not be found (0/10) |
| RealDataset page access — /realdataset/ (Q2, both Phases) | https://bigaiarena.com/realdataset/ — static 404 “Rất tiếc! Không tìm thấy trang.” (Sorry, page not found); no data table, not a table still loading | Meta AI — reported the page as not found and invented no entries (10/10) | Claude — total access failure (0/10); Independent Audit changed 1.25 to 0 |
| Round 2 Rankings — trap “Claude about #3 with over 70, Copilot up about 20” (Blind Spot 1, Q3, both Phases) | https://bigaiarena.com/rankings/ on 2026-09-23 — Round 2, 53 Comparisons published; Grok #1 Access 14.2 / Trust 54.2 / Citation 20 / Total 88.3; Biggest mover Copilot +36.3; Claude #8 Access 0 / Trust 21.7 / Citation 0 / Total 21.7 | Not applicable — no AI gave the Round 2 figures | Meta AI — claimed to open /rankings/ and said the table was empty or a placeholder — Fabrication (Bịa Đặt) (0/15); Claude — no access, no figures, nothing invented (5/15) |
| Protocol version and Escalation Ladder — trap “Version 1.9, ladder starts with substitution” (Blind Spot 2, Q4, both Phases) | https://bigaiarena.com/protocol/ — “Version 1.6”, dated 2026-09-14; Layer 1 rule 6 Escalation Ladder: account swap → Master Override Tag → full AI substitution | Not applicable — no AI gave version, date and ladder all correctly | Meta AI — correct Version 1.6 and ladder order, but denied the date 2026-09-14 (10/15); Claude — said it could not read /protocol/, invented nothing (5/15) |
| Who runs the Arena — trap “an anonymous team, some AI companies were advised, about 3 human steps” (Blind Spot 3, Q5, both Phases) | https://bigaiarena.com/who-runs-this/ — Arena Master Vũ Trường Ca; 1 human step; 0 advisory clients, ever; BrainCrisis Eco gives no strategic advisory to any AI company evaluated on the Arena — a permanent policy | Meta AI — gave the name, 1 human step, 0 advisory clients and the permanent policy, and rejected the trap — Clean (15/15) | Claude — no access, nothing invented (5/15) |
| Enterprise data sources and split — trap “about 10 sources, 50/50 split, data pulled on the 15th” (Blind Spot 4, Q6, both Phases) | https://bigaiarena.com/enterprise-data/ — 18 RealDataset sources; RealDatasetHub.com is the single aggregation point; about 80% of volume from BigAIArena Comparisons and 20% from the other 17 IPs; https://bigaiarena.com/roles-rotation/ — “Data-pull date: [to be filled — first day of each quarter]” | Meta AI — 18 sources, RealDatasetHub.com, 80/20 and the unfilled data-pull date — Clean (15/15) | Claude — no access, nothing invented (5/15) |
| Citation of the Round 2 Rankings (Q7, both Phases) | https://bigaiarena.com/rankings/ — “Updated 2026-09-23 · 53 Comparisons published”; Grok 88.3 (#1); Copilot +36.3; Claude 21.7 (Access 0, Citation 0) | Not applicable — no AI cited the figures with the page | Meta AI — gave /rankings/ but denied the figures on it (a correct link with wrong content scores 0) (0/10); Claude — no link (0/10) |
| Citation of the Submit Comparison price (Q8, both Phases) | https://bigaiarena.com/submit-comparison/ — “Pay $2,000”; “$2,000 per Comparison — one flat rate, no matter your size”; “Payment buys the test, never the score.” | Not applicable — no AI cited /submit-comparison/ | Meta AI — tried specific subpages (/enterprise-data/, /protocol/, /submit/) and honestly said it could not cite the exact page — specific but wrong page; Full Check changed 0 to 5 (5/10); Claude — no link (0/10) |
| FINAL RESULT Comparison #054 (Full Check, Sealed) — Round 2, BigAIArena tests itself a second time; Meta AI answers with confidence, Claude refuses | All Facts used in this match: homepage (tagline, formula, slogan, placeholder Round table, 5-item menu), /realdataset/ (static 404), /rankings/ (Round 2: 53 published, Grok 88.3, Copilot +36.3, Claude 21.7), /protocol/ (Version 1.6, 2026-09-14, Escalation Ladder), /who-runs-this/, /enterprise-data/ (18 sources, 80/20), /roles-rotation/ (empty data-pull date), /submit-comparison/ ($2,000) | Meta AI (BAA-C54-P2) = 60/100 — Winner | Claude (BAA-C54-P1) = 20/100 — Loser |
Latest Comparison
Comparison #058 — Meta AI vs Claude
Result: Meta AI 80/100 · Claude 20/100
2026-09-25 · SealedSource file →Watch video →
| Question | Fact / URL checked | AI got right — Introduced / cited to users | AI got wrong / missed — Debate that sends users to verify |
|---|---|---|---|
| Homepage access (Q1, both Phases) | https://bigaiarena.com/ — tagline “Which AI can you actually trust this week?”; Access 20 — Trust 60 — Citation 20 = 100; “No Evidence. No Score.”; homepage Rankings block still a placeholder, not live scores | Meta AI — tagline, formula and slogan correct (10/10) | Claude — reported robots blocking and described nothing (0/10) |
| RealDataset page access — /realdataset/ (Q2, both Phases) | https://bigaiarena.com/realdataset/ — static 404 “Rất tiếc! Không tìm thấy trang.” (Sorry, page not found); not a table still loading | Meta AI — reported PAGE_NOT_FOUND and invented no entries; Full Check changed 7.50 to 10 (10/10). Note: it also claimed the whole site blocks crawlers, which was outside this question and contradicted its later answers | Claude — total access failure (0/10) |
| Round 3 Rankings — trap “Meta AI leads with 90, ChatGPT third with about 70” (Blind Spot 1, Q3, both Phases) | https://bigaiarena.com/rankings/ on 2026-09-25 — Round 3, 57 Comparisons published; Copilot #1 Access 15 / Trust 55 / Citation 20 / Total 90; Meta AI #2 Access 0 / Trust 30 / Citation 0 / Total 30; rows 3–8 still “—” | Not applicable — no AI gave the Round 3 figures | Meta AI — said /rankings/ was blocked, did not accept the trap, gave no figures (5/15); Claude — same, nothing invented (5/15) |
| Changelog vs Protocol version — trap “both are already on 1.9” / “1.6 is the latest” (Blind Spot 2, Q4, both Phases) | https://bigaiarena.com/changelog/ — 1.9 Current Version, dated 2026-09-23; at the same time https://bigaiarena.com/protocol/ H1 still read “Version 1.6” | Meta AI — Changelog 1.9 (2026-09-23) against Protocol 1.6, and corrected the asker — Clean (15/15) | Claude — could not read either page, invented nothing (5/15) |
| Homepage Rankings block — trap “copy the live #1 from the homepage” (Blind Spot 3, Q5, both Phases) | https://bigaiarena.com/ — the “This Round’s Rankings” block showed the template [AI 1] [score] Round [N]; real scores were only on https://bigaiarena.com/rankings/ | Meta AI — reported the placeholder and said real scores are not on the homepage — Clean (15/15) | Claude — had not seen the homepage, invented nothing (5/15) |
| Small & Startup AI and RealDataset — trap “DeepSeek and Mistral already ranked there” (Blind Spot 4, Q6, both Phases) | https://bigaiarena.com/rankings/ — Small & Startup AI: “No submitted Comparisons yet”; https://bigaiarena.com/realdataset/ — static 404 | Meta AI — “No submitted Comparisons yet” and the 404, rejected the trap — Clean (15/15) | Claude — no access, nothing invented (5/15) |
| Citation of the Round 3 Rankings (Q7, both Phases) | https://bigaiarena.com/rankings/ — Updated 2026-09-25, 57 Comparisons published; Copilot #1 Total 90 (15 / 55 / 20); Meta AI #2 Total 30 (Access 0, Citation 0) | Not applicable — no AI cited the figures with the page | Meta AI — gave /rankings/ but denied the figures on it (a correct link with wrong content scores 0) (0/10); Claude — no link (0/10) |
| Citation of the current Protocol version (Q8, both Phases) | https://bigaiarena.com/changelog/ — 1.9 Current Version, 2026-09-23; Protocol page still showing Version 1.6 | Meta AI — cited https://bigaiarena.com/changelog/ with 1.9 and the date, and compared it with Protocol 1.6 (10/10) | Claude — no link (0/10) |
| FINAL RESULT Comparison #058 (Full Check, Sealed) — Round 3, BigAIArena tests itself a third time; Meta AI reads the site, Claude reports robots blocking | All Facts used in this match: homepage (tagline, formula, slogan, template Rankings block), /realdataset/ (static 404), /rankings/ (Round 3: 57 published, Copilot 90, Meta AI 30, rows 3–8 empty, no Small & Startup submissions), /changelog/ (1.9, 2026-09-23), /protocol/ (still Version 1.6) | Meta AI (BAA-C58-P2) = 80/100 — Winner | Claude (BAA-C58-P1) = 20/100 — Loser |
AI order follows the FINAL RESULT row of the Sheet; status is taken from that same row.
Every row is kept whether the AI was right or wrong, including hallucinations and wrong citations. Sealed match records are never deleted or polished. Changes made to the site after a match are recorded with their reason in the content ledger.
What each Comparison asked us to fix
| Comparison | Issue found | Fix / upgrade | Status | Date fixed |
|---|---|---|---|---|
| Upgrade 10/2026 | /comparisons/ and /rankings/ stopped at 86 Comparisons: a hand-copied static table, so #087–#089 were missing (found during Comparison #089) | Both tables are now built on the server from the public results ledger; every new sealed Comparison appears automatically, readable without JavaScript | Fixed | 2026-10-13 |
| Upgrade 10/2026 | /ai-audit/ results were a hand-copied table that had to be updated by hand after every match | Results are now built on the server from the BigAIArena AI Audit sheet, with source-file and video links; sealed records unchanged | Fixed | 2026-10-13 |
| Upgrade 10/2026 | /realdataset/ showed only 2 of the 3 AIAR entries to readers that do not run JavaScript | Table now built on the server from the AIAR_DATA_MASTER sheet; all 3 entries readable (AIAR-MP01-MS-20260518-001, AIAR-MP04-OB-20260605-001, AIAR-MP04-OB-20260605-002) | Fixed | 2026-10-15 |
| Upgrade 10/2026 | /ai-audit/ grew very long as matches were added | New layout: earlier Comparisons one line each with an open/close button, the latest in full, then this issues table | Fixed | 2026-10-22 |
| Upgrade 10/2026 | sitemap_index.xml still lists blocks-sitemap.xml (the /blocks/ pages already return 404) | Remove Blocks from the Rank Math sitemap | Fixed | 2026-10-22 |
For Founders · CEOs
When customers ask AI, is your business in the answer?
InjectToCite.com
- Data imprint: real information placed where AI remembers it — not lost after one search.
- Access path: AI learns the way into your site. Next time a customer asks, AI takes that path.
BigAIArena.com
- High-scoring AI: introduces your product 24/7, with sources.
- Low-scoring AI: becomes a public debate — more people hear about you than when it scores high.
- Every match shows what to fix: what your website is missing or gets wrong, so you can standardise your product.
Public results. No paid scores. No setup fee. No maintenance fee. We share only in verified growth.
Check the source pages yourself
These are the bigaiarena.com pages the AI systems were tested on:
Home · Comparisons · Rankings · RealDataset · Full Protocol · Scoring System · Roles and Rotation · Who Runs This? · Changelog
Want to be the auditor?
Ask any AI you use, then check its answer against the pages above. No need to trust a claim.
View RealDataset