Google Pomelli Public Beta: Run This Two-Week Split Test for Faster First Drafts

Pomelli goes public beta, and the race to faster first drafts begins

Google Labs and Google DeepMind introduced Pomelli, an AI tool that turns a company’s live website into on brand campaign assets in minutes. For growth and marketing teams, the question is not whether AI can write a post, it is whether Pomelli cuts time to first draft while holding engagement. The credible way to answer that question is a short, controlled experiment.

On October 28, 2025, Google introduced Pomelli as a public beta for English speaking users in the United States, Canada, Australia, and New Zealand. For feature scope and regions, see the official post in Google announces Pomelli. Trade press summarized the launch setup in this public beta availability details.

What Pomelli actually does differently

Pomelli uses a three step workflow designed for non technical marketers:

1. Build a Business DNA profile

• You enter a website URL. Pomelli crawls public pages and images to infer tone of voice, color palette, fonts, and visual style. It codifies this as a Business DNA profile that grounds every generation.

2. Generate tailored campaign ideas

• Pomelli proposes campaign concepts matched to your business profile, or you can seed it with your own idea. Treat it as an ideation accelerator when the team is short on fresh angles.

3. Produce editable, on brand assets

• The tool outputs collections of social posts, ad concepts, banner variants, and copy blocks that you can edit, download, and publish using your usual channels. There is no direct publishing integration, so you keep control of what goes live.

The strategy is simple: automate the heavy lift to the first workable draft, keep humans in the loop for judgment and approvals, and shrink iteration time across formats.

Availability, pricing, and scope at launch

Pomelli launched in public beta on October 28, 2025 and is currently available in English in the United States, Canada, Australia, and New Zealand. Pricing has not been announced for general availability, and during the experimental period access is free. Google describes the product as an early experiment, so expect fast iteration and occasional rough edges.

If you are evaluating broader Google led initiatives, see our take on the WPP and Google AI marketing pact and how to structure pilots around them.

The right way to evaluate Pomelli: a two week split test

Rather than swapping your entire process overnight, run a controlled two week split test focused on two questions: Does Pomelli reduce time to first draft, and do the assets it helps you produce hold or improve engagement?

Here is a concrete plan you can apply this month.

Week 0 setup, keep it simple and rigorous

Define scope and success criteria before you touch the tool.

• Channel scope: pick two channels you can measure cleanly, for example Instagram and LinkedIn, or Google Display and TikTok. Keep the mix stable for the two week run.
• Content types: choose two repeatable formats, for example product announcement post and feature tip carousel. Consistency makes analysis easier.
• Sample size: target at least 20 matched pairs of posts or ads, 10 per week, split evenly across channels and formats. If your cadence is higher, scale up.
• Matching strategy: for each planned idea, prepare a pair of comparable briefs, one for control and one for Pomelli, matched on topic, audience, and intended call to action.
• Success metrics: define primary and secondary metrics now. Primary should be time to first draft (TTFD) and engagement rate by impression. Secondary can include click through rate, save rate, 3 second view rate, and conversion rate for ads where applicable.
• Guardrails: set brand, legal, and compliance rules. List banned claims, required disclaimers, and sensitivity flags the team must check in every draft.
• Roles: assign one producer for time tracking and one editor for brand and legal checks. Use a shared sheet to capture timestamps and revisions.

For adjacent testing ideas in paid channels, review our guide to the Google Ads AssetGenerationService pilot.

Week 1 control, run your current process as usual

For the first week, produce your matched set using your existing workflow without Pomelli.

• Briefing: write the content brief as you normally do. Include audience, single minded message, value proof, call to action, and format specs.
• Production: track TTFD from brief handoff to first editable draft. Capture editing time to first publishable draft. Note how many review cycles were required.
• Publishing: ship the control posts or ads on the planned channels. Record the go live time so you can align windows and account for dayparting.

Week 2 Pomelli, generate first drafts with AI

In the second week, let Pomelli generate the first draft for the matched set, then run your normal editorial process.

• Seeding Pomelli: paste your site URL to build Business DNA. For each brief, use a simple prompt structure. Example: “Create a LinkedIn carousel, audience is sales operations leaders at mid market SaaS, goal is demo requests, tone is clear and no fluff, include a data point and a customer quote placeholder, color scheme must match site, CTA is ‘Book a demo’.”
• Asset selection: ask Pomelli for two to three variants per item and pick one. Resist the urge to hand tweak before you record TTFD. After you log TTFD, make edits as needed for brand and legal.
• Version control: save the raw Pomelli draft and the edited draft. Capture time spent editing and the number of review cycles to reach approval.
• Publishing: ship the Pomelli backed assets on the matching channels with similar timing and budget. Keep your spend and audience targeting comparable for paid tests.

For another channel focused framework, see our TikTok Symphony Automation split test.

What to measure, and how to read the signals

Focus on a small set of metrics you can trust.

• Time to first draft (TTFD): minutes from brief to first editable draft. This is Pomelli’s core promise.
• Edit time to publishable: minutes from first draft to approved asset. This reflects the quality of the AI output and how well it matches your brand.
• Review cycles: number of passes through editor and legal.
• Engagement rate by impression: likes, comments, shares, saves, clicks divided by impressions. Normalize by channel.
• Down funnel proxy: if you can, track click through rate to a landing page and a light conversion, such as email signup or demo request.
• Brand consistency score: a quick 1 to 5 editor rating for voice and visual alignment. Codify examples of 1 and 5 before you start to avoid drift.

Interpretation rules that keep you from overfitting:

• If Pomelli reduces TTFD by 40 percent or more and engagement is within plus or minus 5 percent of control, standardize Pomelli for first drafts in that content type and channel.
• If TTFD reduction is 20 to 40 percent and engagement is within plus or minus 10 percent, continue a longer pilot, expand prompts, and refine editing checklists.
• If engagement drops more than 10 percent, restrict Pomelli to ideation and visual exploration while you iterate on prompts and guardrails.

You can run a simple significance check by grouping pairs and comparing means with a paired test. Do not let low power scare you. Look for consistent directional improvement across pairs and channels.

A sample prompt kit to speed week 2

Strong prompts reduce edits. Treat these as starting points and adapt to your voice.

• Product announcement post, LinkedIn: “Generate a 6 slide carousel that announces our new integration with HubSpot. Audience is B2B marketing managers. Use confident but plain language, include one stat about time saved, avoid buzzwords, headline under 60 characters per slide, CTA is ‘See the integration’.”
• Feature tip, Instagram: “Create two square image concepts plus captions that teach how to use our AI assisted reporting. Voice is friendly and specific, include a micro tutorial in 3 steps, hashtags no more than five, keep captions under 120 words.”
• Paid social concept, Facebook: “Produce three ad concepts with primary text under 125 characters, headline under 40 characters, and image that shows before and after of our analytics dashboard. Audience is SMB owners in retail. Emphasize simplicity, no exaggerated claims.”
• Email header image: “Design a simple banner for a product update email, include our brand colors and logo placement in the top right, leave ample white space for copy, export in 1200 x 400.”

Guardrails and governance, avoid surprises

Any AI tool that ingests your website and outputs copy deserves a few non negotiables.

• Data scope: ensure the site you point Pomelli at has only content you are comfortable using as training context. Remove outdated or experimental pages.
• Claims control: maintain a list of substantiated claims and supporting sources. Require editors to cross check any data point before publishing.
• Legal review: embed legal or compliance review for regulated industries. Ban sensitive topics and make sure the team knows them in advance.
• Image rights: verify that all images generated or remixed for campaigns meet your licensing standards. Keep proof on file.
• Regional limits: if your team sits outside supported regions, do not use workarounds that violate terms. Schedule the test with staff in supported countries.

Where Pomelli likely fits in your stack

Think of Pomelli as a brand informed first draft engine that plays well with the tools you already use.

• Strategy and briefs stay where they are, for example in Notion or a project management tool.
• First drafts come from Pomelli, tuned by your Business DNA.
• Editing and approvals live in your document or design system. Some teams use Upcite.ai to centralize citations and editorial rules so editors can approve faster while keeping an audit trail.
• Publishing flows through your social scheduler, ad manager, or content management system. Pomelli does not publish for you, which is a feature if you care about final sign off.

What good looks like after two weeks

Based on similar AI adoption pilots, you should expect patterns rather than miracles.

• TTFD drops sharply for visual plus short copy formats. Carousel posts, promo banners, and simple ad concepts often see 40 to 60 percent faster first drafts.
• Edit time may stay flat in week 1 of the pilot, then drop in week 2 as editors learn what to accept and how to prompt for the right tone.
• Engagement tends to hold when the message and call to action are clear. It improves when the team uses the time savings to ship more timely and relevant concepts.
• The biggest uplift is usually in the number of viable variants you can test, not in the performance of any single asset.

A lightweight decision framework to standardize, or not

At the end of week 2, score each content type and channel on four questions and make a call.

1. Speed: Did Pomelli cut TTFD by at least 30 percent?
2. Quality: Did engagement stay within 10 percent of control, or improve?
3. Effort: Did total production time fall once you include edits and approvals?
4. Risk: Did any legal or brand issues surface that would require additional safeguards?

Decisions by score:

• 4 out of 4 yes: standardize Pomelli as the default first draft source for that content type. Update your playbook and train more creators.
• 3 out of 4 yes: continue the pilot, refine prompts, add editor checklists, and revisit in two more weeks.
• 2 or fewer yes: keep Pomelli in ideation and visual exploration only. Re visit once the tool or your inputs improve.

Practical pitfalls to avoid

• Prompt sprawl: define two or three reusable prompt templates and stick to them. Random prompts make your analysis noisy.
• Over editing: log TTFD before you touch the draft. Otherwise you will undercount time savings.
• Mismatched pairs: ensure your control and Pomelli pairs are truly comparable. Same topic, same audience, similar timing, similar spend.
• Vanity metrics: track engagement normalized by impressions, not raw likes.
• One and done: run at least 20 matched pairs. Smaller samples are too fragile to guide a policy change.

Beyond the pilot, how to scale responsibly

Once you standardize Pomelli for certain formats, codify the workflow.

• Update your creative brief template to include a Pomelli prompt block and a Business DNA note.
• Create acceptance criteria for drafts, including voice, message clarity, factual accuracy, and visual alignment.
• Store approved assets and prompts in a shared library so future creators can start from proven patterns.
• Train editors on fast triage, for example a three minute pass for tone and claims, a five minute pass for structure and clarity, and a final sign off.
• Track monthly rollups of TTFD, edit time, and engagement so you can see drift and intervene quickly.

The bottom line

Pomelli is a time leverage tool built around your brand signals. The promise is speed to a good first draft, which is often the bottleneck in busy teams. With a two week split test, you will know exactly where it helps and where it does not. Keep the test tight, measure what matters, and make a clear decision by content type and channel. If you standardize Pomelli for first drafts in the right places, you free your team to invest more time in strategy, narrative, and creative ideas that separate you from the pack. If you decide to wait, you still gain a structured playbook for evaluating the next wave of AI tools, and you will be able to apply it quickly.

Run the test, learn fast, and update your playbook. That is how teams turn AI hype into durable marketing velocity.