ClaudeModels & limitsLevel 1: Basic prompts

What Can't Claude Know, and How Do You Fill the Gap?

Claude's built-in knowledge stops at a cutoff, and Anthropic lists June 2026 as the reliable knowledge cutoff for Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5. It also can't see your private files or live data unless you share them or connect a tool. To fill the gap, turn on web search or paste the source.

Prompt to copy 
Question: [your question].

Before answering, sort it into three parts:
1. What you can answer from general knowledge.
2. What may have changed since your training, such as prices, rules, versions or availability. Search for these, even if you feel sure, and cite a source for each one.
3. What depends on my private information. List exactly what you need from me.

If you can't confirm something, say so and tell me where to check it.
  1. June 2026Reliable knowledge cutoff
  2. November 2026About five months laterthis stretch needs search or you
Source: Anthropic's models overview, checked 6 November 2026. The cutoff is the same for Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5.
Claude can answerfrom training or what you share
  • Facts up to about June 2026
  • Reasoning, language, code
  • Anything you paste in
Needs search or youdon't accept it from memory
  • Events after the cutoff
  • Today's prices and rules
  • Your private files and data
A rule of thumb, not an Anthropic chart. The June 2026 date is the reliable knowledge cutoff for current models.

Try it now

  1. 1In a chat, click + in the lower left of the chat window and select Web search, then look for the checkmark. In the new Claude experience there's no toggle, and Claude decides when to search, so ask it directly.
  2. 2On Team and Enterprise, an owner must first enable it at Organization settings > Capabilities.
  3. 3Paste the prompt above with a question about something that changed recently.
  4. 4You'll know it worked when each current fact has a citation. Open at least one and confirm it says what Claude claims. Anthropic notes that some links may not work.

Use it when

When an answer depends on something that changes: prices, laws, software versions, who holds a job, or whether a product still exists. Also use it when the answer depends on your own data, such as this quarter's numbers.

Skip it when

Stable knowledge, such as how compound interest works or what a SQL join does. Web search counts toward your usage limits, and Anthropic's help page suggests asking Claude not to search when you don't need current information.

Why it works

Anthropic defines the reliable cutoff as the date through which a model's knowledge is most extensive and reliable. As of November 2026, that leaves about five months not covered from memory. The harder problem is that Claude doesn't always notice. Anthropic's Sonnet 5.5 guide says the model sometimes answers from training when a web search would catch something that has changed, such as what's allowed, required or charged. So an answer can sound current without having been checked. Once Claude searches or reads your document, it's working from that material. Anthropic's help page says every response that uses web search includes citations, which gives you something to open.

Example

Say you ask whether a vendor's API still has a free tier. From memory, Claude may describe the pricing as it stood before its cutoff. With the prompt above, it searches, cites the vendor's pricing page, and tells you it can't confirm your account's contract terms, so it asks you to paste them. You then know which part came from the web and which part is still open.

Going further

Rule of thumb, not a vendor figure: add a short line to Instructions for Claude telling it to search for prices, rules and versions, so you don't have to ask each time. Developers should check their system prompts too. Anthropic's Sonnet 5.5 guide says wording such as "minimize tool calls" discourages search, and suggests telling the model to check changeable specifics even when it feels confident. Older models you can still select may have different cutoffs, so check the model's own page.

Common mistakes

  • Taking a current-sounding answer as a current one. With no citation, it probably came from memory.
  • Asking Claude what a plan costs or which limits apply to you. Check the pricing page or Settings > Usage.
  • Assuming Claude can see your files because they're on your computer or in a connected app you haven't turned on for this chat.
  • Accepting a citation without opening it. A link shows where to look. It doesn't prove the page says what the sentence says.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Best practice Which Claude Model Should You Use? Mistake Why Does Claude Sound Sure When It's Wrong? Best practice How Do You Stop Claude Inventing API Fields? Technique How Do You Summarize a Very Long PDF With Claude?

Beyond the course

Glossary Large language model Pricing guide Claude pricing, plans and models