Krom Automation includes three AI-powered actions in the free plugin: AI Generate Text, AI Content Moderation, and AI Auto-Tag. These actions connect to external AI language models to generate content, analyse text, and apply taxonomy terms automatically.

This page covers how to set up your AI provider, which models are available, and practical guidance on using each action well.

Setting up an AI provider #

All three AI actions require at least one API key from a supported AI provider. You configure this in Krom Automation > Settings > AI Settings.

Krom Automation supports three providers:

Provider	Models available	Get your API key
OpenAI	GPT-4o Mini, GPT-4o	platform.openai.com
Google Gemini	Gemini 2.0 Flash, Gemini 2.5 Flash Preview, Gemini 2.5 Pro Preview	aistudio.google.com
Groq	Llama 3.3 70B, Llama 3.1 8B, Mixtral 8x7B	console.groq.com

You only need one provider configured to use all three AI actions. If you configure multiple providers, the Auto model option picks the best available model based on which keys are present.

Steps to add an API key:

1. Go to Krom Automation > Settings and open the AI Settings section.
2. Enter your API key for the provider you’ve chosen.
3. Save the settings.
4. Test the connection by running a workflow simulation that includes an AI action.

API keys are stored in your WordPress database and never exposed in the workflow editor or execution logs.

Choosing a model #

Each AI action lets you choose which model to use. Here’s a practical guide to choosing the right one.

Auto is the recommended default for most workflows. It selects the best available model from whichever providers you’ve configured. If you have both OpenAI and Gemini keys, Auto uses the stronger model for the task.

GPT-4o Mini is a fast, cost-efficient OpenAI model. Good for simple text tasks like short email drafts, excerpt generation, and basic summarisation.

GPT-4o is OpenAI’s most capable model in this set. Use it for complex writing tasks, nuanced content moderation, or when output quality matters most.

Gemini 2.0 Flash is Google’s fast and efficient model. Good balance of speed and quality for most automation tasks.

Gemini 2.5 Flash Preview and Gemini 2.5 Pro Preview are Google’s more capable models, suited to longer or more complex prompts.

Llama 3.3 70B is Groq’s most capable open model in this set. Good for text generation where you prefer open-source models.

Llama 3.1 8B and Mixtral 8x7B are Groq’s faster, lighter options. Suitable for high-volume, simpler tasks where speed matters more than output complexity.

AI Generate Text #

This action sends a prompt to your configured AI model and stores the returned text as a workflow variable. You reference that variable in later steps using a merge tag.

Writing effective prompts

The quality of the output depends almost entirely on the quality of your prompt. A few principles that consistently produce better results:

Be specific about the output format. Instead of asking the AI to “write a summary”, tell it exactly what you want: “Write a 1-sentence summary of this post suitable for use as a meta description. Return only the sentence, no labels or explanations.”

Use merge tags to inject real content. Your prompt can include any trigger data. For example, to summarise a post, use {{post_content}} inside the prompt field. The AI receives the actual post content at execution time.

Use the system prompt for persistent instructions. The system prompt field is for instructions that should always apply, regardless of the main prompt. “You are a concise copywriter. Always write in British English. Never use bullet points.” That way your main prompt can focus on the specific task.

Set max_tokens appropriately. For a short excerpt, 100 tokens is plenty. For a full email draft, 500 to 800. For a long-form content piece, up to 4096. Setting it much higher than needed increases cost and response time.

Choose creativity to match the task. Use precise for factual summaries, moderation analysis, and structured output. Use balanced for most general writing. Use creative for social media copy, taglines, and content where variety is valuable.

Using AI output in later steps

Set a clear output_key name and reference it as {{your_key_name}} in any action that follows. For example:

• Set output_key to post_excerpt
• In the next Update Post Field action, set the value field to {{post_excerpt}}
• Krom Automation writes the AI-generated text directly into the post’s excerpt field

You can also use AI output inside a Send Email message body, a Add Post Meta value, an HTTP Request body payload, or any other field that accepts merge tags.

AI Content Moderation #

This action is designed for comment moderation workflows. Pair it with the Comment Submitted trigger and it runs immediately every time a new comment is posted.

How the moderation decision works

The action sends the comment text to the AI and receives a moderation decision: spam, toxic, or clean. Depending on your status_action setting, Krom Automation then automatically applies that decision to the comment.

sensitivity controls how aggressively the AI flags content.

• low: Only flags obvious spam and clearly abusive content. Fewer false positives but may miss borderline cases.
• medium: The recommended default for most sites. Balanced between catching bad content and avoiding false positives.
• high: Flags anything borderline. Good for sites with strict content standards, but expect more legitimate comments to be held for review.

A practical moderation workflow

A reliable comment moderation workflow looks like this:

1. Trigger: Comment Submitted
2. Action: AI Content Moderation with moderation_type = all, sensitivity = medium, status_action = hold, add_label = on
3. Action: Send Email to {{admin_email}} notifying of the held comment

With this setup, every new comment is automatically analysed. Suspicious comments are held for review and the admin gets a notification. Clean comments pass through to WordPress’s normal approval flow.

If you want to be more aggressive, set status_action to spam for confirmed spam and use a condition node to branch based on the moderation result — hold borderline content, spam clear violations, approve everything else.

What this action does not do

The AI Content Moderation action analyses text. It does not access the commenter’s IP address, browser fingerprint, or any external spam database. It’s not a replacement for a dedicated spam filtering plugin like Akismet — it’s an additional layer that’s particularly effective at catching toxic and abusive language that traditional spam filters miss.

AI Auto-Tag #

This action analyses post content and applies matching taxonomy terms automatically. Use it with the Post Published trigger to tag every new post the moment it goes live.

Existing terms vs. new terms

The use_existing setting controls whether the AI can create new taxonomy terms.

• existing_only: The AI matches post content to terms that already exist in your WordPress database. It will not create new tags or categories. This is the safer option for sites with established taxonomies where you want consistent tagging.
• create_new: The AI can create new tags or categories if no existing match is close enough. Useful on new sites or for importing large volumes of content where your taxonomy is still growing.

Getting good tagging results

The AI analyses the content you give it and returns terms that best describe that content. A few things that improve tagging accuracy:

Use title_content as the content source rather than title or content alone. The combination gives the AI more context and produces more relevant terms.

Set max_tags to a reasonable number. Five is a good default for most blogs. Setting it to 15 on a post about a single topic will produce low-relevance terms padded in to fill the quota.

If your taxonomy uses very specific or technical terms that don’t appear naturally in post content (product SKUs, internal codes, industry jargon), use existing_only so the AI matches to what you’ve defined rather than generating generic alternatives.

Troubleshooting AI actions #

The AI action step is failing in the execution log

The most common cause is a missing or invalid API key. Go to Krom Automation > Settings > AI Settings, confirm your key is entered correctly, and check that the provider account is active and has available credits or quota.

The AI output is empty or very short

Check the max_tokens setting. If it’s set very low (under 50), the AI may not have enough room to produce useful output. Also check the prompt — an ambiguous prompt can cause the model to produce a minimal or empty response.

Merge tags in the prompt aren’t resolving

Run the workflow simulator first and check what the merge tags actually resolved to before the AI step. If a tag like {{post_content}} is empty, the post had no content to pass to the AI.

The AI auto-tag is applying irrelevant terms

Try switching content_source from content_only to title_content. Also reduce max_tags — a lower cap forces the AI to return only its most confident matches.