“AI can write a fluent A+ module in 90 seconds. The brand cannot tell whether it sounds like them until they read it out loud. By then it has already shipped.”
The a plus content workflow we run on every A+ rebuild is structured around one binding constraint. AI is never allowed to own the brand voice. Voice is the thing the LLM cannot fake reliably. It is the thing customers register subconsciously.
A polished A+ module that does not sound like the brand reads cleaner than what was there before. It performs worse on conversion. It looks “fine” until the brand owner reads it out loud and realizes it sounds like nobody they know.
The a plus content workflow has eight steps. AI is used in exactly one of them. Downstream of the strategy and the voice fingerprint. Upstream of the human redirect and the designer build. The shape is the same as the broader hybrid stack. AI compiles. Humans own the strategic and judgment work. Applied to the creative function.
Step 1 (human), creative brief from the catalog strategist
Before the creative team touches an A+ rebuild, the catalog strategist produces a brief. The brief contains the keyword tier assignment from the catalog AI/human workflow. The conversion-rate baseline on the existing detail page. The customer cohort the rebuild is targeting. The competitor A+ that ranks well on the category root. The Cosmo-attribute alignment requirements for the category.
This brief is the input to everything downstream in the a plus content workflow. Without it, the creative team is working without a target. AI cannot generate without a target either.
The most common failure mode in the ai-only Amazon agency model is that the AI listing-builder skips this step. The LLM generates A+ content from the ASIN URL and the niche tag. None of the upstream strategic work the brief contains.
Step 2 (human), brand voice fingerprint
The creative lead pulls the brand’s voice fingerprint. They source it from the brand’s existing internal communications. Prior listings the brand has shipped that the brand owner has explicitly approved. The brand owner’s own writing. Emails. LinkedIn posts. Podcast transcripts if available.
The fingerprint is documented in a one-page reference. Sentence length. Vocabulary level. Hedging tolerance. Humor frequency. The words the brand uses and the words the brand explicitly does not use.
This is the artifact the a plus content workflow is built around. Every AI compile pass downstream is constrained by the fingerprint. The strategist team has audited about 40 brand voice fingerprints across the client book. The fingerprints are different enough that an LLM trained on the average of all of them produces text that matches none of them well.
Step 3 (AI compile), first-draft module copy and alt text
With the brief and the voice fingerprint loaded in context, Claude produces first-draft text for each A+ module. Module headers. Module body copy. Image alt text. Comparison-chart row labels.
The instructions are explicit. Stay inside the fingerprint. Hit the Tier 1 keywords naturally. Do not invent product claims. Flag any module where the brief does not give enough specificity to write the copy faithfully.
This is the one step where AI does material content production in the a plus content workflow. The compile saves the creative lead roughly 6 to 10 hours per A+ rebuild on what would otherwise be the slowest part of the work. The compile output is never the final copy. It is the first draft the creative lead redirects in Step 4.
Step 4 (human), creative lead redirect
The creative lead reads the AI’s first draft against the voice fingerprint and against the strategic brief. Roughly 80% of the AI output gets edited at this step. The edits cluster in five categories:
- Voice tightening, the AI’s sentences are usually slightly longer than the brand’s. The vocabulary level usually drifts higher.
- Claim discipline, the AI sometimes writes claims the brand has not made elsewhere. The creative lead removes anything not on the brand’s pre-approved claim list.
- Keyword integration, the AI often hits keywords too literally with verbatim phrase insertion. The lead rewrites to integrate keywords into natural prose.
- Module hierarchy, the AI does not know which module is “above the fold” on mobile versus desktop. The lead reorders the modules accordingly.
- Brand-specific vocabulary, every brand has 4 to 8 words it uses with specific meaning. The AI does not know those words. The lead substitutes them in.
This 80% edit rate is higher than the equivalent rate in the ads workflow. Strategists there edit roughly 70% of AI-compiled output. The difference is that voice is binding in creative work in a way bid clustering is not in ads work. The creative lead is editing for the part of the work AI cannot do well.
Step 5 (human), designer build
The redirected copy goes to a designer for build. The designer composes the modules in Amazon’s A+ content templates. Places the images. Sets the type. Applies the brand’s color and typography system. Produces the comp for brand approval.
AI does not do this step in the a plus content workflow. We have tested AI image generation for A+ content extensively. The current state of the technology produces images that look serviceable on first glance. They fail two of the auto-fail conditions from our brand-assets v3 rules in roughly 60% of generations. Faked text artifacts. Off-brand color. And most commonly, inconsistent stylistic register across the three to five images a single A+ section requires.
The designer’s hand produces a consistent register. The AI’s hand does not. For the same reason, photography for A+ modules is sourced from real product photography sessions the brand has commissioned. Not from AI image generation. The brand approval rate on real photography is roughly 95%. The brand approval rate on AI-generated product imagery, when we tested it, was below 30%.
Step 6 (human), brand approval
The designer’s comp goes to the brand owner (or the brand’s marketing lead) for approval. The a plus content workflow requires brand sign-off before anything ships to Seller Central. No exceptions.
This is the constraint the ai-only Amazon agency model is structurally unable to maintain. There is no human at the agency end of the relationship who can stand behind the work.
Brand approval typically takes one revision cycle. The brand owner flags two or three items. A tone shift. A photo crop. A comparison-chart row to add. The designer revises. The second comp ships about 92% of the time.
Step 7 (rule engine), Cosmo-attribute, character-count, hierarchy checks
Before the approved A+ content is published to Seller Central, the listing_rules module in the ClearSight portal runs three deterministic checks against the content:
- Cosmo-attribute alignment, does the A+ content reinforce the back-end attributes Amazon is indexing the listing against? If the back-end says “sweetener: xylitol” but the A+ headline says “sugar-free,” the check flags a softness in the alignment.
- Character-count compliance, every Amazon A+ module has documented character limits that vary by template. The rule engine validates each module against the current spec.
- Visual hierarchy, comparison charts have row limits. Image-with-text modules have aspect-ratio constraints. Module ordering affects what the customer sees above the mobile fold.
Anything that fails the rules goes back to the designer for a fix before the content ships. This catches the cosmetic problems that would otherwise show up after launch.
Step 8 (rule engine), post-launch CTR and CR monitoring
After the A+ content is live, the listing_rules module monitors click-through rate from search to the detail page and conversion rate on the detail page itself. On the 7-day and 30-day post-launch windows. If either metric is below the baseline by more than a defined threshold, the rule engine flags the rebuild for review. The creative lead investigates.
This closes the loop on the a plus content workflow. The same rule engine that ran the pre-launch checks also tracks whether the rebuild produced the conversion lift the strategy targeted.
Across the trailing 12 months of A+ rebuilds on the book, the conversion lift on the parent ASIN runs about +11% at 30 days post-launch. With a tail that reaches +18% by 90 days as the listing fully indexes. The 11% number is the median. The distribution skews higher for brands where the prior A+ content was AI-generated by the previous agency. Lower for brands where the prior A+ was already professionally produced.
What AI does and does not do in the a plus content workflow
AI is used in exactly one of the eight steps. Step 3 (first-draft module copy and alt text). The rest of the workflow is human strategy. Human voice work. Human design. Human approval. And rule-engine validation.
The shape is deliberate. Brand voice is the part of creative work where AI generates output that looks fine and fails on read-aloud. Keeping AI out of the voice-binding steps is the structural reason this workflow produces A+ content that converts.
The stack-reference post describes the broader architecture this fits into. The next post in this series moves to inventory forecasting. The workflow shape changes again because the binding constraint is math accuracy rather than brand voice.
Reviewed by the Catalog Team and the SEO Team.
The Hybrid Stack, 10-post series. You are reading post 8 of 10.
Black-hat track, three archetypes of the ai-only Amazon agency:
- The three black-hat shapes of the 2026 Amazon agency
- Why offshore VA + ChatGPT shops are the most expensive cheap option
- Why AI listing-builder SaaS can’t get a 7-figure brand to actually index
- Why AI-first consultants who never log into Seller Central miss the work that moves money
White-hat track, the ClearSight hybrid stack:
- The ClearSight intelligence layer, stack reference
- Catalog AI/human workflow
- Ads AI/human workflow
- Creative AI/human workflow (you are here)
- Inventory AI/human workflow
- 12 months of the hybrid stack, results recap
← Previous: Ads AI/human workflow | Next: Inventory AI/human workflow →