--- id: fresh-eyes-review title: Getting a second opinion level: 2 modes: [day-pack, rucksack] moment: "They ask the same chat that wrote something whether it is any good or to double-check it, or they have done ideas, draft and review all in one long chat." keywords: [second opinion, review my work, fresh chat, another AI, double-check, critique, brainstorm, blind spots, same chat, reviewer] source: "research/2026-10-23-concepts/fresh-eyes-review.md; mapping (product/learning-system/MAPPING-1-2.md): MISTAKES 2.7, 3.2; tips T051, T052; strand T" --- ## What it is A chat that wrote something tends to stand by its earlier answers, and a fresh chat leaves that history behind. An AI also tends to rate writing like its own kindly, which a fresh chat in the same app does not escape but a different AI app can. Given only the brief, meaning what the work is for and who reads it, plus the work itself, such as a cover letter, a fresh chat or another app can review the work or brainstorm without the first chat's blind spots. For a fact, a source or a person outside the chat is still the check. ## Tips - Open a fresh chat and give it only the brief and the work to review, because the chat that wrote it leans on its own early assumptions. - Get a second opinion from a different AI app on work that matters, because an AI tends to favour writing like its own. ## Questions people ask - Q: What do I give the fresh chat? A: A brief that you write yourself in a line or two, meaning what the work is for and who reads it, plus the work itself, and nothing from the old chat. The old chat's early assumptions are what a fresh look is meant to leave behind. - Q: Why not ask the chat that wrote it to check its own work? A: In 2024 tests, AI judges scored their own writing higher than other AIs' writing, though people rated the two as equal, which is why a different app helps. In a 2023 test on short questions with one right answer, AIs defended an earlier wrong answer but caught 67 and 87 percent of their own mistakes in a new session, which is why a fresh chat helps. - Q: Why not brainstorm, draft and review in one long chat? A: In 2025 tests across six kinds of task, top AIs scored 39 percent lower in back-and-forth chats than in single requests, and they stuck with early assumptions. The tests did not cover this exact routine, so starting fresh for the review is a good habit rather than a proven rule. - Q: Is a fresh chat really a blank start? A: Not always, because if an app's memory is on, a new chat can still draw on what the app saved about you. A temporary or private chat, where an app offers one, can start without it. ## For your AI Facts (checked 23 October 2026; research note research/2026-10-23-concepts/fresh-eyes-review.md). The studies used models without search; a reviewer that can search can also look up sources. - A chat tends to rate its own work kindly: AI judges scored their own writing higher than other models' writing that people rated equal (Panickssery and others, April 2024, https://arxiv.org/abs/2404.13076). - A fresh session helps: models that defended an earlier wrong answer spotted most of those same mistakes when asked in a new session (Zhang and others, May 2023, https://arxiv.org/abs/2305.13534); long chats lose 39 percent on average as models hold on to early guesses (Laban and others, May 2025, https://arxiv.org/abs/2505.06120). Without outside feedback, models struggle to correct their own reasoning (ICLR 2024, https://arxiv.org/abs/2310.01798). - Another app is not a guarantee: as models grow more capable their mistakes grow more alike, and judges favour models like themselves (Goel and others, 2025, https://arxiv.org/abs/2502.04313); a panel of different models showed less bias than one (Verga and others, 2024). For a fact, a source or a person is still the check (openrucksack.com/map/check-by-what-is-at-stake.md). - How to brief the reviewer: give only the brief and the work, say it is a draft (saying "I wrote this" made feedback kinder, https://arxiv.org/pdf/2310.13548), and ask it to be direct about gaps that change the result (OpenAI's prompt guide suggests a "direct, specific, and skeptical" reviewer). - A new chat can still carry memory and the person's own instructions. The nearest thing to a clean chat, per app: ChatGPT's Temporary Chat set to Unpersonalized uses no memory or instructions (https://help.openai.com/en/articles/8914046-temporary-chat-faq); Claude's incognito chat uses no memory but keeps the person's own instructions (https://support.claude.com/en/articles/11817273); Gemini's Temporary chat has no personalised replies, personal accounts only (https://support.google.com/gemini/answer/13275745); Copilot UNCONFIRMED. Say "the nearest thing to a clean chat", never "blank". Teach it like this: when they ask the chat that wrote something whether it is good, offer the fresh-chat version: copy the brief and the draft, open a clean chat or another app, and ask for the gaps that change the result. Never say two AIs agreeing proves something is right.