Last verified: September 2, 2026
What You Can Now Do
Context Memo's Page Builder copilot chat now accepts messages up to 20,000 characters, up from the previous 2,000-character cap. That's the difference between pasting one long paragraph and pasting a full source article, a positioning brief, or a substantial chunk of a call transcript. The decision it changes is simple: you no longer have to pre-summarize source material before handing it to the copilot. Paste the whole thing. The specifics that survive (named customers, dates, measured figures) are exactly the sentences AI models pull into answers.
Where It Is in Context Memo
The change is in the copilot chat inside the Page Builder. Open any memo in the builder and the chat panel is where it has always been. Nothing about the interface moved. The only difference is the size of what the input box will take.
How to Use It
- Open the memo you're working on in the Page Builder and click into the copilot chat panel.
- Paste your full source document in a single message: the whole article, the whole brief, or the transcript section you care about. Don't trim it first. The copilot reads the entire message as one block of context.
- Add your instruction in the same message, above or below the pasted source. Tell the copilot what you want built from it: a section, a comparison, a set of claims with the proof attached.
- Review the draft the copilot returns and check that the specifics carried through. Named customers, dates, and figures from your source should appear in the output, not generalized paraphrases of them.
- Refine in follow-up messages using short, targeted instructions. Because the full source is already in context, follow-ups can reference details from anywhere in the document rather than only the last thing you pasted.
- Publish the memo to your domain when the draft holds up. Schema markup and citation structure work the same as before.
Why We Built It
The old limit created a hidden quality tax. If a source document ran 12,000 characters and the box took 2,000, you had three bad options: summarize it yourself, split it across messages and hope the copilot stitched the context back together, or paste a fragment and let the model fill in the gaps. All three degrade the output.
Self-summarizing is the worst of the three. Summarization is lossy in exactly the wrong direction: it keeps the general claims and drops the specifics. A sentence with a named customer, a date, or a measured figure gets cited. A sentence that says "companies see strong results" does not.
Multi-message pasting fails differently. Context degrades across turns, and the copilot weights the most recent message heaviest. The tail of your source gets more attention than the opening. You end up with thin framing and an overweighted close.
What It Does Not Do Yet
20,000 characters is a larger window, not an unlimited one. A long call transcript can still exceed it, and in that case you'll need to select the section that matters rather than paste the raw export end to end. We'd suggest choosing the segment where the customer describes the problem in their own language, since that's the material that produces publishable copy.
The copilot also doesn't ingest files. Paste is still the input method, so PDFs, docs, and recordings need to be converted to text on your side first.
One more thing worth being clear about. Marketing leads and founders at B2B software companies keep raising the same complaint to us: generic meeting summaries are too raw to publish, and the editorial work required to turn a transcript into an on-brand asset is what actually blocks the work. A bigger input box removes one constraint in that chain. It does not do the editorial judgment for you. You still decide which claims are defensible, which quotes are usable, and what your brand will stand behind in a published memo.