AI Post & Image Generation

The platform's AI does two things for content production: write text (via Anthropic Claude) and generate matching images (via Google Gemini). Both are wired into the content calendar, review responses, and intake-form review flows.

Review sentiment analysis runs on Claude Haiku 4.5 (not Gemini). Image generation stays on Gemini.

How AI generation works (under the hood, briefly)

For any AI generation, the platform assembles a context bundle:

1. Brand profile — voice, audience, things to mention, things to avoid (see Brand Profiles).
2. Content quality guidelines — org-wide rules (see Shared Libraries).
3. Business context — industry, location, recent posts.
4. Your prompt — what specifically you want for this post.
5. (For images) — color palette from the brand profile, style references from your library.

Then it calls Claude (text) and Gemini (image) and returns drafts.

Text generation

Where it's available

• Content calendar editor → Generate with AI.
• Series generation.
• Review response panel → Generate Reply.
• Report narrative sections → Draft monthly summary.
• Intake response auto-summary.

Patterns that produce good output

• Specific prompts beat vague ones. "Spring service special for HVAC clients in Phoenix highlighting AC maintenance" beats "make a post."
• Reference recent posts — the platform auto-includes the last 5 published posts so the AI doesn't repeat itself.
• State the call-to-action. "Ending with a call to book a free estimate" produces a clear CTA every time.
• Iterate on tone. If output is too corporate, write "make it more conversational" as a follow-up — the editor supports back-and-forth.

Variations

Today the platform returns a single draft per generation. A "generate N variations in parallel and pick your favorite" UI, plus a per-org Settings → AI Usage page for tracking generations against your monthly cap, are on the Roadmap.

Image generation

Where it's available

• Content calendar editor → image area → Generate.
• Profile media tab → bulk image generation for upcoming posts.

What works

• Natural lighting, real-life settings. Avoid prompt language that asks for "professional studio."
• No text in images. The platform enforces a "never include text in images" instruction at both Claude and Gemini prompt layers because Google deranks text-heavy images.
• People are tricky. Gemini can hallucinate strange faces — prefer environment shots or stylized hands-in-work shots.
• Brand colors via brand profile. The platform feeds the brand profile's primary color into Gemini as a soft constraint.

Watermarking

If the brand profile has a watermark image set (with position, shape, and background color configured), the platform applies the watermark after Gemini generation, during image post-processing. See Brand Profiles.

Image post-processing

After Gemini returns the raw image, the platform:

1. Crops to 4:3 (1200×900 — GBP's preferred ratio).
2. Adjusts contrast/saturation lightly for consistency.
3. Applies watermark if configured.
4. Stores in Google Cloud Storage.
5. Returns the public URL.

Costs and quota

Each text generation and each image generation count as one AI call. Your plan has a monthly cap; the platform enforces it server-side. A user-facing Settings → AI Usage page (showing current month's consumption + history), a per-profile "Disable AI generation" toggle, and a per-generation audit log (who triggered it, what prompt, what output, which model + version) are all on the Roadmap.

Overage packs available — see Plans & Billing.

---

Next: Approval Workflows