--- title: "Restart a long AI conversation before its answers get worse" description: "AI answers degrade as a conversation grows. Learn why context rot happens, which symptoms to watch, and how to restart without losing your decisions." canonical: "https://scalewithsearch.com/articles/why-does-ai-get-worse-in-long-conversations" date: "2026-01-28" modified: "2026-09-25" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # Restart a long AI conversation before its answers get worse. For the first 10 messages, the AI is excellent. It understands the project, asks good questions, and produces what you need. By message 30, it suggests options you already rejected. It asks questions you already answered. It contradicts what it said at message 15. You are still in the same conversation, and nothing changed except its length. The drop is not random. As a conversation grows, the model uses its earlier parts less well. This page explains why, lists the symptoms, and gives a restart procedure that keeps your decisions. ## Know that attention is uneven A language model uses an attention mechanism to process your messages. For each answer, it reads everything in the context window and weights some parts more than others. That weighting is not even. A 2023 study found that models use information at the start and end of a long input more reliably than information in the middle (Liu et al., "[Lost in the Middle](https://arxiv.org/abs/2307.03172)"). In a 40-message conversation, messages 15 to 25 sit in that weaker middle zone. The model has not deleted those messages. They are still in the context window. It uses them less. As the conversation grows, earlier context shapes fewer of the answers. ## Know what context rot means Context rot is the name for a drop in model performance as the input grows. More context feels like it should help. Past a point, it makes answers worse. In July 2025, Chroma published a report on the effect (Chroma, "[Context Rot](https://www.trychroma.com/research/context-rot)"). It tested 18 models and found that performance fell as input length grew, even on simple tasks. Models with very large windows showed the same pattern. A larger window lets a model hold more text. It does not make the model use all of that text equally well. Cost adds pressure. In the standard transformer design, the work to process the context grows quadratically with its length. Twice the context means about four times the work. Products respond in different ways. Some drop the oldest messages near the limit. Some summarize them. Claude Code, for example, compacts a long session into a summary when you run `/compact`, and it does so automatically near the limit. Each approach keeps the session running and loses some detail. The effect applies to every product: ChatGPT, Claude, Gemini, and Copilot. The problem is attention, not storage. ## Check for the four symptoms Watch for these signs in a long conversation: - **Contradiction.** At message 35 the AI suggests an approach it rejected at message 12. The earlier decision sits in the weak middle of the context. - **Repeat questions.** You gave your constraints at message 8. At message 30 it asks for them again. - **Falling quality.** Early answers are detailed and specific. Late answers are vague, generic, or off target. - **Ignored corrections.** You corrected an error at message 20. By message 35, the error is back. Two or more symptoms mean the conversation has passed its useful length. Restart it with the procedure below. ## Know why one endless conversation makes it worse Some people avoid new chats. They keep one conversation open for days or weeks, on the theory that the AI then has more to work with. The effect is the reverse. The longer the conversation, the more the product drops or compresses, and the less the model uses the early parts. By message 50, much of your effort goes into working around the AI instead of with it. ## Know why the usual workarounds fall short People try four fixes: - **Frequent restarts.** Some users start a new chat every 10 to 15 messages. It works, but each restart loses the thread, and you explain the project again. - **Summaries.** Some users summarize the first 20 messages and paste the summary into a new chat. The summary keeps what you decided and loses why. If you choose the wrong details to keep, the model cannot recover them. - **Memory features.** ChatGPT and Copilot can save facts that you ask them to remember. They keep "prefers Python," not the design discussion behind it. Saved facts also do not combine into a task on their own. - **References to message numbers.** "As I explained in message 10" makes the model look back some of the time. You still do the memory work. Each fix makes you the context manager. You decide what the AI keeps, what to summarize, and when to restart. That is overhead on top of the work. ## Move the lasting context into a file The durable fix is to separate two kinds of context. Lasting context is who you are, what you build, how you work, what you tried, and what you rejected. Working context is the task in front of you. Put lasting context in a Markdown file. Keep the conversation for the current work. A tool such as Claude Code reads a file named `CLAUDE.md` from your project folder at the start of each session. The file goes in before the first message, so it sits at the start of the context, a strong position. It is not immune to long-chat effects, so keep it short and focused. The [CLAUDE.md template for business context](/articles/claude-md-template-business-context) shows a compact structure. In the ChatGPT app, a file does not load from a folder by itself. You upload it or add it to a Project. The principle is the same: the settled context comes from a maintained file, not from message 12 of an old thread. A file has four advantages over a long thread: - **It reloads in full.** Each new session reads the current file from the start. Nothing builds up to compress. - **It stays constant.** A conversation degrades as it grows. The file changes only when you edit it. - **It sits at the start.** The model reads it before your first message, in the zone where recall is strongest. - **You control it.** You decide what the AI must know. If a rule stops working, you rewrite the line instead of guessing which message the model ignored. ## Restart with a handoff file A file makes a restart cheap. Use this procedure when the symptoms appear or after a major decision point. 1. Ask the AI to list the settled decisions, rejected options, constraints, and open questions from the conversation. 2. Check each item against your own notes and the source documents. 3. Correct any item the AI got wrong. 4. Add the checked items to your project file under dated headings. 5. Start a new session with the updated file loaded. 6. Ask the new session to restate the constraints and the rejected options. 7. Compare its answer with the file, then continue the work. Do not accept the AI's list without review. A summary written by a degraded conversation carries the same gaps as the conversation. Step 2 is the step that matters. A handoff entry can be short: ```markdown ## 2026.09.25 pricing page decisions - Decided: three tiers, annual billing only - Rejected: usage-based pricing (support cost too high) - Constraint: no discount above 15 percent - Open: launch date, pending the trade show schedule ``` The [ChatGPT same-chat guide](/articles/why-chatgpt-forgets-same-chat) gives a fuller checkpoint format for long contract and scope work. If a stale value survives the restart, the [context reset procedure](/articles/reset-ai-context-without-deleting-memory) helps you find which layer still holds it. ## Keep long work, not long threads Short conversations are not the goal. Deep, repeated work on one project is where AI earns its place, and that work can run for weeks. The goal is to keep the lasting state out of the thread. When settled decisions live in a file, a restart costs a minute instead of an hour. The thread can then stay as long as the current task needs, and no longer. ## Related: AI memory - [Choose an AI for long conversations that still recalls the first hour](/articles/best-ai-for-long-conversations) - [Find out why AI lost a detail mid-conversation and restore it](/articles/why-does-ai-lose-context-mid-conversation) - [Budget an AI context window so critical information stays in view](/articles/ai-context-window-how-ai-loses-information) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```