how to use Jev to check your ai marketing team's work before you review it jev is a new model built to evaluate information and make structured decisions inside software you give it context and specific questions, and it returns choices, scores and probabilities. that makes it useful for checking your marketing agents' work before it reaches you for review when your agents create a landing page, social posts or an email sequence, you can have the work checked against the campaign brief and your requirements an eval is a check against a requirement you've defined. for marketing, that could mean checking for ai tells, unsupported claims or confusing copy here's how to set that up: 1. agree on the requirements with your lead agent start with your campaign goal, brief and the assets you're creating have the agent ask for your experience, previous feedback and anything you already know needs attention. include the mistakes you keep correcting when reviewing work for a landing page, define checks for: > ai tells in the copy, like forced contrasts, repetitive phrasing and generic filler > invented facts, quotes, customer results or unsupported claims > whether product details and other information match your approved sources > confusing headings, unclear buttons or missing information someone needs to take the next step > visual issues like cramped spacing, weak contrast, inconsistent typography and broken mobile layouts give the evaluator examples of work you've approved and rejected, with your reasons. turn "this looks generic" into specific things it can check save those requirements in the campaign brain before the creating agent starts work jev can assess the copy and supplied evidence. pair it with browser tests and visual inspection for the ux/ui checks, then send the combined findings back to the creating agent before your review 2. give the evaluation step the right context your company brain holds reusable knowledge about the business, including positioning, voice, supported claims and exclusions the campaign brain holds the brief, research, decisions and working files for the campaign have your workflow retrieve the relevant files and send them to jev alongside the asset and evaluation questions if you're checking a customer result, include the case study that supports it. for writing style, include your banned patterns and examples with your edit notes have the workflow request missing material before evaluating the draft 3. ask specific questions jev can evaluate jev takes context and questions with defined answer formats it can choose from options you provide, score something against criteria or estimate the probability that a statement is true. independent questions can run in parallel against the same context for your landing page, that could look like: > does this paragraph use any of the banned writing patterns? > does the case study support the result quoted in this paragraph? > do the product details match the supplied documentation? > does the button label explain what happens when someone clicks it? your workflow uses those answers to decide what happens next, and the creating agent handles any revisions 4. send failed checks back with enough context to fix them your orchestrator coordinates the work and assigns tasks to the agents connect the evaluation results to its revision process: > failed checks return to the agent responsible for the asset > uncertain results go to your review queue with the issue flagged > assets that pass reach your sign-off queue with their evaluation attached if a customer claim fails, send the copy agent the specific claim, the failed requirement and the supporting material it was evaluated against after the revision, run the checks on the updated asset again set a limit on revision attempts. anything that keeps failing should reach you with the unresolved issue and what the agent has already tried 5. include technical and visual checks you can use this process across your marketing work, with different tools handling different checks code can check character limits, required fields and tracking parameters. browser tests can check forms, links and how a landing page renders designs and videos need tools that can inspect their visual content, along with your review. jev evaluates the information you supply, so confirm which formats it supports before assigning it checks on those assets keep the results attached to the asset, including anything that wasn't checked also run checks as work progresses. research should be evaluated before it informs the brief, and claims should be checked before they get reused across the campaign 6. test the checks against work you've already reviewed take previous assets you've approved and rejected, including your edit notes run the evaluation against them and compare its decisions with yours look closely at drafts that pass even though you would reject them. you may need to make a requirement more specific, provide missing context or change how the question is asked jev's confidence estimates also need testing against your examples before they control whether work continues automatically keep the results advisory during testing, with a person reviewing the decisions 7. save what you change during review when an asset reaches you, review the work alongside its failed checks, revisions and anything still unresolved you still decide whether the angle makes sense, whether the creative feels right and whether it's ready to publish save the reasons for your edits in the campaign brain if you keep correcting the same issue, turn it into a proposed evaluation requirement. test it against previous assets before adding it to the workflow keep campaign-specific feedback with that campaign