Troubleshooting Quick-Reference Checklist

A fast lookup for the everyday Claude.ai issues covered across this section: match your symptom to a likely cause and a quick fix, then follow the link if you need the full walkthrough.

How to Use This Checklist

Scan the Symptom column for what you're actually seeing, not what you assume is causing it.
Try the Fast Fix first; it's the quickest thing worth attempting before reading the full article.
Follow the linked article when the fast fix doesn't resolve it, or when you want the complete checklist and reasoning behind it.
Bookmark this page as the entry point back into the rest of the section.

Symptom to Fix Map

Answers feel vaguer, slower, or less relevant than earlier in the same conversation.
- Likely cause: Context bloat or a stale thread building up as the conversation grows.
- Fast fix: Concisely restate your current goal and any constraints in one message rather than continuing to react to the drifted thread.
- Full article: Why Claude Answers Get Worse: A Mental Model
You get a direct length-limit message, or replies start getting truncated or oddly cut short.
- Likely cause: The conversation has reached, or is approaching, its maximum length.
- Fast fix: Start a fresh conversation and open it with a short restatement of the goal, key decisions, and current state, not a full transcript.
- Full article: Fixing Hit Context and Length Limits in Long Conversations
A citation, URL, quote, or source looks off, or you just can't find it when you search.
- Likely cause: A fabricated citation, generated as a plausible-sounding continuation rather than a verified reference.
- Fast fix: Search for the source independently before relying on it, and ask Claude directly whether it's certain or inferring.
- Full article: What to Do When Claude Hallucinates Citations or Sources
You open a new chat, or switch devices, and Claude has no memory of your prior conversation.
- Likely cause: Separate conversations don't share memory by default, only conversations inside the same Project do.
- Fast fix: Open with a concise restatement covering the goal, decisions already made, and where things currently stand, then verify Claude picked it up correctly before continuing.
- Full article: Walkthrough: Recovering a Conversation That Lost Context
Headings, bullets, bolding, or numbering look inconsistent between replies.
- Likely cause: Formatting is generated fresh per response, and can drift over a longer or more ambiguous thread, or the paste destination isn't rendering markdown.
- Fast fix: State the exact format you want explicitly, and check whether the same answer renders correctly in a markdown-aware app before assuming the output itself was wrong.
- Full article: Fixing Inconsistent Formatting in Claude's Responses
You're not sure whether to start a new chat or use a Project for something you'll come back to.
- Likely cause: The task's recurrence, or lack of it, hasn't been weighed against Project setup overhead.
- Fast fix: Default to a plain chat for one-off work; move to a Project once you've manually restated the same background more than once or twice for the same ongoing topic.
- Full article: Projects vs Plain Chats: Choosing the Right One
You suddenly can't send a new message at all, with a usage or limit notice, and it's not about conversation length.
- Likely cause: A rate limit tied to your Free or Pro plan's rolling usage window.
- Fast fix: Read the reset time in the in-app notice, wait it out or consolidate remaining questions into fewer messages, and consider a lighter model like Claude Haiku 4.5 for simple tasks in the meantime.
- Full article: Walkthrough: Diagnosing Rate Limits on Lower Plan Tiers

Applying This Checklist in Order

Start with the mental model (item 1) whenever the symptom is fuzzy, general quality decline is the umbrella most of the other, more specific symptoms sit under.
Move to the specific-symptom entries (items 2-7) once you can point to a concrete, distinct sign, a limit message, a bad citation, a lost-context chat, formatting drift, an unclear Project decision, or a blocked send.
Treat items 2 and 7 as the two to tell apart first if you're simply blocked from continuing: item 2 is about one conversation's length, item 7 is about your plan's usage cap. The exact notice wording tells you which.

FAQs

Where should I start if I'm not sure what's actually wrong?

Start with item 1, the general answer-quality entry. Most of the more specific symptoms in this checklist are variations on the same underlying context dynamic.
If you can point to something concrete, like a specific error message, jump straight to the matching entry instead.

What's the difference between item 2 (length limit) and item 7 (rate limit)?

A length limit is about a single conversation reaching its maximum size, it only happens in long conversations. A rate limit is about your plan tier's usage cap and can happen even in a short, brand-new conversation.
The specific wording of whatever notice you see is the reliable way to tell them apart.

Do I need to read the full articles, or is this page enough?

This page is meant for a fast first pass. If the fast fix resolves your issue, you may not need the full article at all.
If it doesn't, or you want the complete checklist and reasoning, the linked article has the full depth.

Why is there no single "basics" article for this section?

This section is organized around specific symptoms rather than a single onboarding flow, since troubleshooting starts from whatever problem you're actually hitting, not from a fixed starting point.
This checklist serves as the closing, tie-everything-together page instead.

Can more than one of these symptoms apply at the same time?

Yes, it's common for a long conversation to show both general quality decline (item 1) and formatting drift (item 5) at once, since both stem from the same growing context.
Address the underlying context issue first; formatting and citation issues often improve once the conversation itself is cleaned up or restarted.

Is a fabricated citation (item 3) related to the other context issues?

It's a related but distinct symptom. Context bloat affects overall answer quality; a fabricated citation is specifically about generating a plausible-looking but unverified reference, which can happen even in a short, fresh conversation.
Treat citation verification as its own habit, not something a fresh conversation alone will fix.

How do I know if I should move to a Project (item 6) instead of just following the lost-context walkthrough (item 4) again?

If you've followed the lost-context recovery steps more than once or twice for what is clearly the same ongoing topic, that repetition is the specific signal to set up a Project instead.
A single multi-day task is usually fine handled with manual restatement; a topic you return to regularly is a stronger fit for a Project.

Does switching Claude models fix any of these symptoms directly?

Not on its own for most of these. Context bloat, length limits, citation habits, formatting drift, and Project decisions are largely independent of which model you're using.
The one exception is quota management under a rate limit (item 7), where deliberately choosing a lighter model like Claude Haiku 4.5 for simple tasks does help conserve usage.

What should I do if none of these seven symptoms match what I'm seeing?

Re-check the exact wording of any error or notice you received, it usually narrows things down to one of the entries above.
If it genuinely doesn't match any of these, it likely falls outside the scope of this everyday-troubleshooting section.

Is this checklist a substitute for reading the mental-model article?

It's a fast lookup, not a substitute. Why Claude Answers Get Worse: A Mental Model explains the underlying mechanism that several of these symptoms share, which is worth understanding if these issues keep recurring.

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.

How to Use This Checklist

Scan the Symptom column for what you're actually seeing, not what you assume is causing it.

Try the Fast Fix first; it's the quickest thing worth attempting before reading the full article.

Follow the linked article when the fast fix doesn't resolve it, or when you want the complete checklist and reasoning behind it.

Bookmark this page as the entry point back into the rest of the section.

Symptom to Fix Map

Answers feel vaguer, slower, or less relevant than earlier in the same conversation.

Likely cause: Context bloat or a stale thread building up as the conversation grows.
Fast fix: Concisely restate your current goal and any constraints in one message rather than continuing to react to the drifted thread.
Full article: Why Claude Answers Get Worse: A Mental Model

You get a direct length-limit message, or replies start getting truncated or oddly cut short.

Likely cause: The conversation has reached, or is approaching, its maximum length.
Fast fix: Start a fresh conversation and open it with a short restatement of the goal, key decisions, and current state, not a full transcript.
Full article: Fixing Hit Context and Length Limits in Long Conversations

A citation, URL, quote, or source looks off, or you just can't find it when you search.

Likely cause: A fabricated citation, generated as a plausible-sounding continuation rather than a verified reference.
Fast fix: Search for the source independently before relying on it, and ask Claude directly whether it's certain or inferring.
Full article: What to Do When Claude Hallucinates Citations or Sources

You open a new chat, or switch devices, and Claude has no memory of your prior conversation.

Likely cause: Separate conversations don't share memory by default, only conversations inside the same Project do.
Fast fix: Open with a concise restatement covering the goal, decisions already made, and where things currently stand, then verify Claude picked it up correctly before continuing.
Full article: Walkthrough: Recovering a Conversation That Lost Context

Headings, bullets, bolding, or numbering look inconsistent between replies.

Likely cause: Formatting is generated fresh per response, and can drift over a longer or more ambiguous thread, or the paste destination isn't rendering markdown.
Fast fix: State the exact format you want explicitly, and check whether the same answer renders correctly in a markdown-aware app before assuming the output itself was wrong.
Full article: Fixing Inconsistent Formatting in Claude's Responses

You're not sure whether to start a new chat or use a Project for something you'll come back to.

Likely cause: The task's recurrence, or lack of it, hasn't been weighed against Project setup overhead.
Fast fix: Default to a plain chat for one-off work; move to a Project once you've manually restated the same background more than once or twice for the same ongoing topic.
Full article: Projects vs Plain Chats: Choosing the Right One

You suddenly can't send a new message at all, with a usage or limit notice, and it's not about conversation length.

Likely cause: A rate limit tied to your Free or Pro plan's rolling usage window.
Fast fix: Read the reset time in the in-app notice, wait it out or consolidate remaining questions into fewer messages, and consider a lighter model like Claude Haiku 4.5 for simple tasks in the meantime.
Full article: Walkthrough: Diagnosing Rate Limits on Lower Plan Tiers

Applying This Checklist in Order

Start with the mental model (item 1) whenever the symptom is fuzzy, general quality decline is the umbrella most of the other, more specific symptoms sit under.

Move to the specific-symptom entries (items 2-7) once you can point to a concrete, distinct sign, a limit message, a bad citation, a lost-context chat, formatting drift, an unclear Project decision, or a blocked send.

Treat items 2 and 7 as the two to tell apart first if you're simply blocked from continuing: item 2 is about one conversation's length, item 7 is about your plan's usage cap. The exact notice wording tells you which.

FAQs

Where should I start if I'm not sure what's actually wrong?

Start with item 1, the general answer-quality entry. Most of the more specific symptoms in this checklist are variations on the same underlying context dynamic.
If you can point to something concrete, like a specific error message, jump straight to the matching entry instead.

What's the difference between item 2 (length limit) and item 7 (rate limit)?

A length limit is about a single conversation reaching its maximum size, it only happens in long conversations. A rate limit is about your plan tier's usage cap and can happen even in a short, brand-new conversation.
The specific wording of whatever notice you see is the reliable way to tell them apart.

Do I need to read the full articles, or is this page enough?

This page is meant for a fast first pass. If the fast fix resolves your issue, you may not need the full article at all.
If it doesn't, or you want the complete checklist and reasoning, the linked article has the full depth.

Why is there no single "basics" article for this section?

This section is organized around specific symptoms rather than a single onboarding flow, since troubleshooting starts from whatever problem you're actually hitting, not from a fixed starting point.
This checklist serves as the closing, tie-everything-together page instead.

Can more than one of these symptoms apply at the same time?

Yes, it's common for a long conversation to show both general quality decline (item 1) and formatting drift (item 5) at once, since both stem from the same growing context.
Address the underlying context issue first; formatting and citation issues often improve once the conversation itself is cleaned up or restarted.

Is a fabricated citation (item 3) related to the other context issues?

It's a related but distinct symptom. Context bloat affects overall answer quality; a fabricated citation is specifically about generating a plausible-looking but unverified reference, which can happen even in a short, fresh conversation.
Treat citation verification as its own habit, not something a fresh conversation alone will fix.

How do I know if I should move to a Project (item 6) instead of just following the lost-context walkthrough (item 4) again?

If you've followed the lost-context recovery steps more than once or twice for what is clearly the same ongoing topic, that repetition is the specific signal to set up a Project instead.
A single multi-day task is usually fine handled with manual restatement; a topic you return to regularly is a stronger fit for a Project.

Does switching Claude models fix any of these symptoms directly?

Not on its own for most of these. Context bloat, length limits, citation habits, formatting drift, and Project decisions are largely independent of which model you're using.
The one exception is quota management under a rate limit (item 7), where deliberately choosing a lighter model like Claude Haiku 4.5 for simple tasks does help conserve usage.

What should I do if none of these seven symptoms match what I'm seeing?

Re-check the exact wording of any error or notice you received, it usually narrows things down to one of the entries above.
If it genuinely doesn't match any of these, it likely falls outside the scope of this everyday-troubleshooting section.

Is this checklist a substitute for reading the mental-model article?

It's a fast lookup, not a substitute. Why Claude Answers Get Worse: A Mental Model explains the underlying mechanism that several of these symptoms share, which is worth understanding if these issues keep recurring.

Troubleshooting Quick-Reference Checklist

Search Documentation

Troubleshooting Quick-Reference Checklist