Prompts tell Renable how to create content for each of your products, from ad copy and headlines to AI-generated lifestyle
images and videos. A new project has no prompts, so nothing is written until you create one. Each one starts from
instructions you can adjust, for example to change the tone of voice.
Prompts live on a project. Open a project, go to its Creative tab, and use the Prompts section to create or open
one. A prompt runs for every catalog in its project. To limit it to some of them, use Use as.
Creating a prompt#
Click Create prompt and pick a Type of prompt to start from. Each type pre-fills instructions and sets a default
content category.
| Type of prompt | What it produces |
|---|
| Write ad copy | Ad copy |
| Write ad copy by looking at the product image | Ad copy, written from the product image rather than existing text |
| Write headlines | Headline |
| Write reason to buys | Reason to buy |
| Write product titles | Product name |
| Write text-to-voice scripts | Voice script |
| Write product highlights (AI discovery) | Short benefit lines, tagged ai-highlights |
| Write enriched descriptions (AI discovery) | A fuller factual description, tagged ai-description |
| Write use cases (AI discovery) | Short "good for" phrases, tagged ai-use-cases |
| Generate lifestyle images | An AI-generated lifestyle image |
| Generate lifestyle videos | An AI-generated video |
| Animate lifestyle images to video | An AI-generated video, using an existing lifestyle image as its source |
| Custom prompt | Ad copy, starting from a blank instruction |
The three AI discovery types write extra product content for Google's AI shopping surfaces. They only describe what
your product data already says, and they're fact-checked. Their output comes tagged, so a channel field can ask for that
content alone, and it doesn't change the copy your other prompts produce. See
AI discovery on Google.
The lifestyle image, lifestyle video and animate types generate media: see
AI image and video generation. A video made by a prompt doesn't reach your ads by
default.
When prompts run#
Prompts run automatically:
- The first time you set up a project, for all its products.
- When new products appear in your feed.
- When a product changes, for example its description, images or price. Any affected channels are updated too.
- When you publish a prompt, new or edited. Every product in its scope is reprocessed within minutes, so publishing an
active prompt spends credits on every product in scope.
To stop a prompt, turn off the Activate prompt toggle and publish. This removes the text, images and videos it
generated from your products, including anything you approved, so whatever used that content falls back or comes up
empty. Narrowing its scope under Use as does the same for the products it no longer covers. Archiving a prompt on the
Creative tab does the same, with no publish step.
Why a prompt skips some products#
A prompt can only write for a product that has the content it asks for. When the field a placeholder reads is empty, or the
prompt asks for a Product image and none of the product's images are processed, the prompt generates nothing for that
product and spends no credits.
Anything the prompt generated for that product earlier is removed, including images you approved. If the prompt later
produces the same image again, it comes back approved.
A product whose image is still downloading or being processed isn't skipped: the prompt waits and generates once the image
is ready. The same goes for a feed image Renable hasn't fetched yet.
The warning on the prompt#
When a prompt skips products, it carries a warning such as Skips 1,204 feed items — Lifestyle headline. You'll find it
in the notes and events at the bottom of the prompt's page, in the dashboard's Needs attention queue, and as the
prompt's status dot on the project's Overview and Creative tab.
The number counts feed items: a feed item is one product in one catalog, so a product sold in three markets is three feed
items. A prompt writes separately for each catalog, so a product can be skipped in one market and written for in another.
See Feed items and products.
The warning names the inputs that are missing, how many feed items each affects, and why, for example:
Product image — 1,180 feed items: None of this product's images are processed: neither the catalog's Image processing settings nor a Process images rule covers them.
Description — 24 feed items: The product has no value for this.
It lists up to five inputs, and ends with what helps for each reason.
| The warning says | What it means |
|---|
| The product has no value for this. | The field the placeholder reads is empty for these products. |
| None of this product's images are processed: neither the catalog's Image processing settings nor a Process images rule covers them. | The prompt asks for a processed image, and the catalog doesn't process these products' images. |
| This product has no Lifestyle image. (or whichever kind the placeholder asks for) | Their images are processed, but none is in that category. |
| Categorize images is off for this product, so none of its images is categorized as a Lifestyle image. | With Categorize images off, every processed image counts as a Product image. |
| This product has no images. | The feed lists no images for these products. |
| The image could not be downloaded. | Renable couldn't fetch the image from your website: see Product Images Won't Download. |
| The image host isn't answering Renable's downloads right now, so the image hasn't been downloaded yet. | The website or CDN hosting these images has stopped answering, and the catalog shows the warning Your image host isn't answering Renable's downloads. See Product Images Won't Download. |
| The product has a matching image, but it was rejected, hidden by a tag, or left out by the placeholder's settings. | An image of the right kind exists, but the placeholder isn't allowed to use it. |
| The source image did not resolve to one of the product's images. | The prompt's source image must be one of the product's images. |
How to fix it#
- Process those images, when the products' images aren't processed. Turn on Categorize images or add a step in the
catalog's Image processing settings, or add a Process images rule for these products. See
Which images are processed. Or add a
fallback to Image link if the feed's own photo will do.
- Add a fallback to the placeholder, when the products have processed images but none the placeholder can use, or when
a value is empty. Click the ⋮ menu at the end of the expression and choose Add fallback. For an image, Product
image, else Image link uses the feed's own photo. Publish the prompt afterwards.
- Fill in the missing value in your source feed, or with a rule on the catalog.
- Let the image download. An image that couldn't be downloaded is tried again automatically, and the prompt generates
once it arrives. If it keeps failing, see Product Images Won't Download.
- Wait for the image host, when it isn't answering. There's nothing to change in Renable: it keeps trying, and the
prompt generates once the image downloads. A fallback to Image link doesn't help, since that photo comes from the same
host.
- Leave those products out of the prompt by narrowing its scope under Use as, for example to a product set without
them. Whatever it generated for products that fall out of scope is removed.
Skipped products are generated for as soon as they have what the prompt needs: after you publish the fix, or when the
missing image or value arrives.
The warning updates itself, and ends once no product is skipped or when you switch the prompt off and publish. An archived
warning stays archived while the number changes, and it can come back archived. Tick Show archived events in the events
filter to see it.
Content generation blocked#
Content generation blocked — Lifestyle headline is a different warning: the prompt can't generate for any product,
because of something in the prompt itself. Its text says what, for example "It is empty for every product, so the prompt
cannot generate for any of them." (an instruction or step with nothing in it) or "This model needs a source image, and
the prompt has none." Fix the prompt, or switch it off, and publish.
The prompt editor#
- Description: a note for your own reference. It doesn't affect what the prompt does.
- Instructions: tell the AI what to do, much as you'd brief a freelance copywriter. Prompts default to English, but you
can write in your own language. Either way, include an instruction to write the output in the product's native language.
- Steps: one or more input steps that supply the product information. Each step holds placeholders such as Title or
Description, filled in per product when the prompt runs. Click the icon on a step to toggle it between Input and
Example response, and use "Add prompt step" and "Delete prompt step" in a step's ... menu to build multi-step
prompts. See Advanced prompting techniques.
- What to avoid: what the model should leave out, such as extra text or the wrong setting. Shown only for video models
that support it.
- Generated content: shows the output as it streams in during a test run.
Fact-checking#
AI can write claims the product's information doesn't support. Turn on Automatically fact-check the results (in
Advanced) to run every output through a second check against the product data before it's accepted. You can customize
the Fact-checking instructions, choose a Fact-checking model, and add a Regular expression to match for falsehoods
(case insensitive) to catch specific phrasing.
Use as#
This accordion controls where the content is used and how it's tagged: which catalogs, projects, product sets, languages,
content categories and tags it applies to. Use it to scope a prompt narrowly, for example to one product set, or to tag its
output so a specific template placeholder or channel field can target it.
Advanced#
- Model: which AI model runs the prompt. Pick a faster, cheaper model for high-volume, straightforward tasks, and a more
capable model when quality matters more than cost.
- Aspect ratio, Resolution, Duration, Looping video: shown only when the selected model generates images or
video. The options depend on the model.
Previewing and generating#
Use the Sample product field to find a product and run the prompt on it before rolling it out. The button
reads Generate while idle and Cancel during a run. For multi-step prompts you can read the input and output of each
step, which helps when the result isn't what you expected.
Once a run completes, Renable shows an estimate of what the prompt will cost in credits across the products in its scope.
Check it before you publish, especially for image and video prompts, which typically cost more per product than text. See
How credits work.