Name the client and their site. Juma returns a claims register where every claim carries the public source that backs it, that source's publication date and the scope the evidence covers, the approved sentence, the phrasing to avoid, and a paste-ready answer block an AI engine can quote and attribute.
When an assistant answers a question about your category, it quotes whoever published something specific, dated and attributable. Most brands lose that moment twice over: the claims they can prove are buried in a results release nobody links to, and the claims on the homepage are the ones no public evidence reaches. Name the client and their site.
Juma reads the brand's own published material, then returns the claims they can defend with the source and publication date beside each one, the approved sentence and the phrasing to avoid, an answer block written for the question people actually ask, and an exposure audit of what is already in market that the evidence does not support. It arrives as a branded PDF plus an editable CSV register with an owner, a first review date and a retirement path per claim. 400+ marketing teams use Juma for work like this, with human review on every output and a named person signing off every claim before it is used.
Most substantiation risk sits in copy that is already live. This step reads the brand's own marketing surfaces and lists the claims being made there that the public evidence does not reach: a dated metric presented as current, a study result generalised past its cohort, a category superlative with nothing behind it. Each entry names what is claimed, why the evidence falls short, and the specific evidence that would close the gap, so the team can either qualify the line, replace it, or commission the proof. It is the fastest route from a claims library to a change in what gets published.
Run the exposure audit. List the claims this brand already publishes in its own marketing that the public evidence does not reach, and for each one give the claim as written, where it appears, why the evidence falls short, the safe qualified version, and the specific evidence that would close the gap.
A claim an assistant can reach is a question heading, a direct answer of roughly forty to seventy words, and a labelled source line carrying the exact source title, its publication date and the canonical URL. That shape is what makes the answer extractable without losing the evidence attached to it. This step writes one block per claim in the brand's voice, ready to paste onto the page that should own it. The source line is the part teams skip and the part that decides whether the quote gets attributed to the brand or to whoever documented the fact better.
Write the paste-ready answer block for every claim: the question a person actually asks an assistant, a 40 to 70 word answer in the brand's voice, and a labelled source and methodology line with the exact source title, publication date and canonical URL. Name the single owned page that should carry each block.
The evidentiary bar moves the moment a brand says more, better, faster or cheaper than someone else, named or implied. This step sorts the register into claims that can carry a comparison, claims that can only be stated about the brand's own performance, and claims that may never be made comparative at all. For the first group it states what the comparison would need: a like-for-like metric, a defined competitor set, a dated measurement and a method someone outside the company could repeat. It is the section that keeps a confident campaign line out of a regulator's inbox.
Sort the register by comparative risk: which claims can carry a comparison to a named or implied competitor, which can only be stated about this brand, and which may never be made comparative. For the first group, state the evidence a defensible comparison would need.
Claims expire at different speeds. Mission language holds for years, a course count or a feature availability line can be wrong in a month, and a daily-active-user figure is out of date the morning after the next earnings release. This step gives each claim a cadence, a first review date counted from the issue date, a named role owner, and a retirement path: who pulls the line, which surfaces it has to come off, and what replaces it in the interim. The standing rule is that a period-specific metric is replaced rather than quietly overwritten, so the brand keeps a record of what it said and when.
Build the review calendar. Give every claim a cadence, a first review date counted from today, a named role owner, and a retirement path naming who pulls the line, the surfaces it comes off, and what replaces it while new evidence is pending.
An answer engine attributes a fact to the page that owns it. This step maps every claim to one canonical destination, mission and access claims to the brand or about page, availability to a maintained catalogue, research results to the efficacy page beside the methodology, financial and ranking facts to the investor archive with the period label intact. It also rates each source for durability, marking whether the evidence sits on a stable page, a dated recap post or a periodic release, and names the permanent page that should carry a claim whose only current source is a post that will be superseded next year.
Map each claim to the single owned page that should carry it, and rate every source for durability: stable canonical page, dated recap post, or periodic release. Where a dated post is the only evidence, name the permanent page that should carry the claim instead.
A Juma Project is a shared space where the team keeps everything Juma needs to know about a client. Create one project per client and every run starts from the same evidence base. For claims work the payoff is direct: the register reuses wording the client's compliance owner already cleared, cites the research the client actually published, and stops proposing language that was rejected last quarter.
Legal-cleared claims and required qualifiers
The claims the client's legal or compliance owner has signed off, with the qualifier each one has to carry. The register then uses cleared wording verbatim instead of writing fresh language that has to go back through review before anything can be published.
Research, studies and methodology notes
Efficacy studies, benchmark reports and any internal research the client has published, with the method summary. This is what lets a result keep its cohort and its limits when it gets turned into a sentence, which is the difference between a defensible claim and a generalisation.
Investor and press facts with their periods
Results releases, the press kit, the company boilerplate and the facts page, each with the period it covers. Financial and scale claims keep their reporting period attached, so a figure from February does not quietly become a statement about today.
Brand voice and claim ownership
How the brand writes, and who owns which category of claim. Answer blocks come back sounding like the client rather than like a compliance memo, and every row in the register names a role that can actually approve it.
Add a short description to each knowledge item in the project's info field so Juma knows what the file holds and when to reach for it. For example:
A citable brand claim is a statement the brand can defend, written in a form an answer engine can lift and attribute. It carries four things: the sentence itself, the public source that backs it, the date that source was published, and the scope the evidence covers. A claim with no source is marketing language. A claim with a source but no date goes stale without anyone noticing, and a claim with no scope gets generalised past what the evidence supports the first time someone shortens it.
The practical test is whether a stranger could check it. If an assistant quotes the claim and a reader follows the citation, they should land on a page published by the brand that says the same thing, in the same period, about the same group. That is the standard the register is built to.
Two files. The branded PDF is the document that goes to a client or a compliance reviewer: the defensible claims with approved phrasing and the wording to avoid, the claim type and substantiation test for each one, the paste-ready answer blocks, the canonical destination map, the comparative claims rule, the exposure audit, and the review calendar with its owners. The editable CSV is the operational register, one row per claim with the source URL, publication date, evidence excerpt, scope limit, destination, owner, approver, approval date and next review.
The split matters in practice. The PDF is what gets read and approved once, and the CSV is what the team maintains for the next year as evidence changes and claims are retired or replaced.
A messaging framework decides what the brand wants to say: the positioning, the pillars, the proof points per audience. A claims library decides what the brand is allowed to say, and in exactly which words. The two sit next to each other, and the register is what keeps the framework honest when a pillar depends on a number that was true two years ago.
The other difference is the audience. A framework is written for marketers. A claims register is written for marketers, legal reviewers and answer engines at the same time, which is why every row carries a source, a scope and an approver rather than a tone note.
No, and any tool promising that is selling something. What the Flow does is remove the reasons a brand gets skipped: facts with no date, claims with no source on an owned page, and answers buried in a release that was never written to be quoted. Publishing a specific, dated, attributable answer on the page that owns the question is what makes a citation possible.
Visibility also has to be measured rather than assumed. Pair this with an audit of how answer engines currently describe the brand, re-run it after the answer blocks are live, and watch which claims start appearing with attribution.
Whoever signs off claims today. The register is built so that review is quick rather than optional: every row names an approver role and an approval date, and the document states plainly that no claim is published without a named human signing it off. Juma assembles the evidence, the phrasing and the limits; the decision to stand behind a claim stays with the people accountable for it.
That is the shape of the whole Flow. It augments the work a brand, legal and content team already do rather than replacing their judgement, and the parts that carry risk are the parts a person is asked to confirm. Strategy, taste and judgment stay human.
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.
