Name the client and the accounts, and every personalised line comes back checked against the page it came from, with the publication date beside it and anything stale dropped before a single email is written.
Name the client, their website, the target accounts and the role to write to, and the research happens first: each account's newsroom, blog and release notes are opened, the publication date is read off the page, and anything older than 90 days is dropped and replaced. What comes back is a source check table showing every URL, its printed date, its age in days and whether it passed.
Only the sources that pass become copy. The sequence runs three touches per account on days 1, 4 and 9, each built on a different source, each written to one named role, with word counts, two subject line variants, a sender identity and an opt-out line. Two files land in the chat: a branded PDF of the full sequence and an editable CSV with one row per email. More than 400 marketing teams use Juma, and outbound is the place where an unchecked detail costs the most, because a prospect who spots a five-month-old launch described as news stops reading at the first line.
Every claim in a cold email carries a date, and the date is where outbound goes wrong. This step opens each page the sequence intends to cite, reads the publication date printed on that page rather than inferring it from the URL or the position in a feed, and measures it against today. Sources outside the window are dropped and replaced. The output is a source check table listing every URL, the date as printed, the age in days, and PASS or DROPPED, so the claim that sources are current is something the table proves rather than something the copy asserts.
Before writing anything, open every page you plan to cite for Canva, Duolingo and Notion, read the publication date printed on the page, and give me the source check table with the age in days and PASS or DROPPED for each one.
Most sequences stretch one announcement across three emails, so the day 4 and day 9 touches restate the day 1 hook in different words. This step gives each touch its own source, so the second email opens on something the first did not mention and the third opens on something neither covered. The day 9 close-out carries a lower-friction ask, a short reply or a one-page resource rather than a third meeting request. Each body is word-capped, 120 for the opener and 75 for the follow-ups, with the count printed so the limit is checkable.
Give each touch in the Notion sequence a different source, and make the day 9 email a close-out with a lower-friction ask instead of another meeting request.
One account list usually supports several plays, because the same company news reads differently to a growth lead, a content lead and a technical buyer. This step takes the sources already verified and rewrites the sequence around a new named role, reframing the ask, the proof and the rep notes to that role's outcome while keeping the same checked facts underneath. The research does not run twice. What changes is who the email is for, which is the part most sequences get wrong when they address a job title list instead of a person.
Rewrite the same three sequences for a Head of Content instead of a VP of Product Marketing, keeping the verified sources and changing the framing, the ask and the rep notes.
The CSV is built to be imported rather than read, with one row per email and real line breaks inside the body cells so the copy arrives intact instead of as escaped characters. Columns cover the account, the role, the send day, both subject variants, the body, the word count, the evidence label, the source URL and its printed date, the opt-out line and the rep notes. This step maps those columns onto the fields your sequencer expects, so the sender identity, the physical address and the suppression list are handled by the platform and the copy drops straight into the steps.
Map the CSV columns onto the fields our sequencer imports, and tell me which compliance items the sending platform has to supply rather than the copy.
Teams build one Juma Project per client and add context over time. Outbound benefits from this more than most jobs, because the research changes every week while the offer, the proof and the rules about who not to contact stay the same. Add them once and every sequence starts from the same footing instead of being re-explained in each prompt.
Offer and Positioning One-Pager
What the client sells, who it is for, and the specific outcome a buyer gets. Without it, the value sentence in every email defaults to a feature list pulled off the website. With it, the ask connects the account's own news to something the client actually does.
Proof Library
Named customers, case study numbers and the results the client can stand behind. These are the claims allowed in the body copy, which keeps the sequence away from the unverifiable statements that make cold email read as noise.
Do Not Contact and Disqualifiers
Current customers, open opportunities, competitors, regions the client does not sell into, and any account under an agreement that forbids outreach. This is the list that prevents the most expensive kind of outbound mistake.
Account List and Role Map
The accounts in play and the named role to write to at each one. With this saved, the sequence starts from the right target instead of a job title guess, and running the play for a second role becomes one line rather than a new brief.
Sequences Already Sent
Past sequences and which sources they used. New runs then avoid opening on a trigger a previous touch already covered, which matters when the same account is worked twice in a quarter.
Add a short description to each knowledge item in the project's info field so the right file gets picked for the right task. For example:
Every opening line is labelled Evidence-backed or Hypothesis, and the two are handled differently. Evidence-backed means the page was opened and the claim is supported by what it says. A Hypothesis never reaches the email body: it moves to the rep notes as something to confirm before sending.
The source check table is the part that makes this checkable rather than a promise. It lists every URL considered, the publication date printed on that page, the age of that source in days, and PASS or DROPPED. A reader can audit the sequence line by line without opening a single tab, and the DROPPED rows show what was rejected and why.
This matters because the standard failure in AI-written outbound is not a bad sentence, it is a confident sentence about something that happened five months ago. Reading the date off the page rather than inferring it from a URL or a feed position is what catches that.
Two files: a branded PDF of the full sequence and an editable CSV with one row per email. The PDF carries the source check table, every email with its evidence label and word count, account-specific rep notes and the compliance guidance, organised by account and then by send day.
The CSV is the working file. Columns cover the account, the role, the send day, both subject line variants, the body, the word count, the evidence label, the source URL, its printed date, the opt-out line and the rep notes. Body cells carry real line breaks, so the copy imports into a sequencer intact.
Hand-written outreach is usually better on voice and worse on coverage. A rep researching three accounts properly spends a morning on it, and in practice the third account gets a thinner version of the first account's email. This Flow does the research pass at the same depth for every account and shows its working, which is where the time goes back.
What it does not do is decide who to contact, what to offer, or whether the angle is right for this relationship. Strategy, taste and judgment stay human. The rep notes exist precisely because some things need a person to confirm them, and the Hypothesis label is an instruction to go and check rather than a suggestion.
Human review on every output is the intended workflow here, not a disclaimer. A sequence that leaves the chat and goes straight to a send list without a rep reading the notes is using half the deliverable.
The copy handles the parts that live in the copy. Every email names the sender and their company and carries a one-line opt-out, and each account gets a short note covering CAN-SPAM and GDPR business-to-business legitimate interest.
The sending platform still has to supply the rest: a valid physical mailing address, suppression list handling, and honouring opt-outs within the required window. The run says so explicitly rather than implying the files are launch-ready on their own. Rules also differ by market, so a legal read before a first campaign into a new region is worth the hour.
Yes, and the honest answer is that the output changes shape. When an account has nothing inside the 90-day window, the run says so plainly and opens on a durable fact about the company instead of dressing up something old as new. That is a weaker email, and it is better than a false one.
For those accounts, widening the window in the prompt is the usual fix: 180 days for companies that publish a few times a year. The source check table still prints the age of every source, so the trade being made stays visible rather than hidden inside the copy.
Name the lead and their URL, and this Flow returns a pre-call account brief where every claim is labeled Confirmed with a source URL or Inferred, plus an ICP fit score and a CRM-ready lead record.
400+ marketing teams trust Juma. ChatGPT, Claude, and Perplexity know why. Ask them.
Ready when you are
Start typing a task and Juma will figure out the rest.
