AI text shortcodes
AI shortcodes put text written by a language model into your posts. Use them in any post template, in the meta title and description, and inside other places. Each shortcode is one request to the AI provider, billed by the provider on your own API key.
Basics
- These shortcodes work only while RSS Retriever builds a post. Typed into a normal WordPress post, they do nothing.
- The
promptis your instruction to the model. Use placeholders to give it the article:%post_content%,%post_content_notags%,%post_title%and so on. - Do not use double quotes inside parameter values; use single quotes:
prompt="Write a title, do not use 'quotes'". - Spintax works in all parameters, so each post can get a slightly different instruction:
prompt="Rewrite this in a {friendly|formal|witty} tone: ...". - Ask for HTML when you need HTML. If a model answers in Markdown anyway, tick Convert Markdown to HTML.
- If a request fails or the key is missing, the post is not added and the reason is written to the log.
[ai_generate] accepts an AI engine name and calls the right provider. Switching from GPT to Claude or to an OpenRouter model is a matter of changing one word.
[ai_generate]
The universal text shortcode. Works with OpenAI, Anthropic, Google, xAI, OpenRouter and your custom endpoints.
[ai_generate engine="anthropic-claude-sonnet-4-6" prompt="Rewrite this article in plain English, keep the HTML structure: %post_content%"]prompt(required)- The instruction.
engine- The AI engine name. Without it, the Default AI model is used.
max_tokens- Maximum length of the answer in tokens.
temperature- Randomness. Default: 1. Lower values give more predictable text, higher values more creative text.
[openai_gpt]
OpenAI GPT models with your OpenAI key.
[openai_gpt model="gpt-4o" prompt="Translate the following article to German and return HTML: %post_content%"]prompt(required)- The instruction.
model- The OpenAI model name, for example
gpt-5.5,gpt-5.4-mini,gpt-4o. Default:gpt-4o-mini. GPT-5 and newer models are called through the OpenAI Responses API with a low reasoning effort suited to writing. max_tokens,temperature,top_p,presence_penalty,frequency_penalty- Optional generation parameters.
top_pis an alternative to temperature (nucleus sampling):0.1means only the top 10% most likely tokens are considered. The penalties (-2.0 to 2.0) discourage repeating words and topics. Change either temperature or top_p, not both.
[claude]
Anthropic Claude models with your Anthropic key.
[claude model="claude-sonnet-4-6" prompt="Translate the following article to Spanish and return HTML: %post_content%"]prompt(required)- The instruction.
model- The Claude model name, for example
claude-opus-4-7,claude-sonnet-4-6orclaude-haiku-4-5. Default:claude-sonnet-4-20250514. max_tokens- Maximum length of the answer. Default: 4096.
temperature,top_p,top_k- Optional.
top_klimits sampling to the K most likely tokens and removes unlikely “long tail” answers.
[gemini]
Google Gemini models with your Google AI key.
[gemini prompt="Write a recipe article for '%post_title%'. Use only p, ul, li and em tags."]prompt(required)- The instruction.
model- The Gemini model name, for example
gemini-3.5-flashorgemini-3.1-pro-preview. Default:gemini-2.5-pro. max_tokens- Maximum length of the answer. Without it, the plugin uses a large limit suited to the model.
temperature,top_p,top_k- Optional generation parameters.
[xai]
xAI Grok models with your xAI key.
[xai model="grok-4.3" prompt="Write a short, punchy news summary of this article: %post_content_notags%"]prompt(required)- The instruction.
model- The Grok model name. Default:
grok-4-latest. max_tokens,temperature,top_p- Optional generation parameters.
[or_text]
Any text model available on OpenRouter: OpenAI, Anthropic, Google, Meta Llama, Mistral, DeepSeek, Qwen and hundreds more, with one OpenRouter key.
[or_text model="mistralai/mistral-large" prompt="Rewrite this text in a sarcastic style, keep its HTML structure: %post_content%"]prompt(required)- The instruction.
model- The OpenRouter model ID, for example
anthropic/claude-sonnet-4.6,openai/gpt-4o,deepseek/deepseek-v4-proormeta-llama/llama-3.1-8b-instruct:free. - Other parameters
- Any parameter of the OpenRouter API is passed through:
max_tokens,temperature,top_p,top_k,presence_penalty,frequency_penalty,repetition_penalty,min_p,top_a,seed(repeatable results for identical input), depending on what the model supports.
[custom_ai]
A model of your own custom AI endpoint: DeepSeek, Mistral, a local server or any OpenAI-compatible API.
[custom_ai id="deepseek" model="deepseek-chat" prompt="Summarize this article in three sentences: %post_content_notags%"]id(required)- The Endpoint ID from the Accounts page.
prompt(required)- The instruction.
model- The model name. Optional if the endpoint has a default model.
- Other parameters
- Passed to the endpoint as they are, for example
max_tokensortemperature.
[gpt_article]
Writes a long, coherent article in several sections. Language models produce limited text per answer; [gpt_article] works around this with a chain of prompts: it asks the model for a list of section titles first, then writes each section, giving the model the article written so far, so the parts connect naturally.
[gpt_article topic="%post_title%" engine="openai-gpt-4o" sections="5" directives="Format the text for a WordPress post using only p, em, blockquote, ul and li tags. Mention known facts and little-known secrets."]topic(required)- The subject of the article, for example
%post_title%orLife in Prague. engine- The AI engine name. Default:
openai-gpt-4o. sections- Number of sections. Default: 4. More sections, longer article. With
1, the article is written in one request without a section list. sectionN- Extra instructions for section N only, added to
directives:section2="Add one quote." structure- Your outline instead of the model’s choice:
structure="Section 1: Introduction; Section 5: FAQ". directives- Formatting and content rules for every section. A sensible default (paragraph tags only) is used when you leave it out.
language- Language of the article. Default: English. Example:
language="Spanish". heading- The heading tag placed before every section. Default:
h2. Useheading=""for no headings. max_tokens- Maximum tokens per section. Default: 4096.
temperature- Creativity. Default: 1.
How structure and directives work together
The section titles are generated from the topic and the structure, for example “make the first section an introduction and the last one a FAQ”. Each section is then written with both structure and directives in its prompt, so the article follows your outline and your formatting rules.
A 5-section article means 6 requests: one for the outline and one per section. The Delay of the feed is applied between sections, which helps with providers that limit requests per minute.
Combining shortcodes
Use the Template processing order to feed the result of one template into another. Write the article first, then a title for it:
Post content: [gpt_article topic="%post_title%" sections="4"]
Post title: [ai_generate prompt="Write a catchy title (no quotes, no colons) for this article: %post_content_notags[300]%"]For short tasks inside other text, the %gpt_prompt[...]% placeholder is often handier than a shortcode.