TYPO3 AI Prompts Library: Take Control of Every AI Instruction

An editor uses an AI feature in TYPO3 to create a meta description. The first result is too long, the second sounds like an advert, and the third ignores the brand’s language rules. The team blames the model, but the instruction behind it may be the real problem.

T3AF brings prompts from connected extensions into one central library in the TYPO3 backend. Teams can review, customise, test and restore instructions instead of repeatedly repairing weak output.

This guide explains how TYPO3 AI prompt customisation helps agencies and B2B teams improve content quality, apply brand rules and build repeatable AI workflows.

The Black Box Problem in AI Tools for TYPO3

Most AI features give editors a simple process: add content, choose a model, generate and review.

What is less visible is the prompt controlling the result.

Editors Can See	Editors May Not See
Source content	Length and format rules
Selected model	Tone and terminology
Generated result	Forbidden language
Generate button	Missing-information rules

A prompt may define the output length, language formality, required structure and claims the AI must avoid. Without access to it, teams cannot easily tell whether poor output comes from the model, the page content, AI Context or the instruction itself.

Why Hidden Prompts Create Business Problems

Unclear prompts can lead to inconsistent metadata, shifts between “du” and “Sie”, unsupported claims and repeated manual corrections.

For agencies, these problems multiply across client projects. Editors regenerate content, reviewers fix the same issues, and similar workflows produce different results.

Prompt Transparency Supports Governance

Visible prompts allow the right teams to correct the instruction:

• SEO specialists can define metadata limits.
• Brand teams can control terminology.
• Localisation teams can set language and formality.
• Compliance reviewers can restrict unsupported claims.

This turns prompt editing into a shared quality process rather than trial and error by individual editors.

When an instruction is hidden, teams can only repair poor output. When it is visible, they can improve the source.

What Is the T3AF AI Prompts Library?

The T3AF AI Prompts library is a central TYPO3 backend module for managing instructions used by connected AI extensions.

AI Foundation provides the shared interface and storage, while connected extensions contribute their prompt categories and feature-specific instructions. These prompts are then loaded when the related AI feature runs.

Administrators can open AI Foundation → AI Prompts, filter the catalogue by extension and review the prompts already available.

When the required prompt exists, they can inspect or customise it. When it is missing, they can create a new prompt by selecting the category and adding its title, description, prompt type and instruction.

What Teams Can Control

From the central library, teams can:

• Filter prompts by connected extension
• Review existing prompt instructions
• Edit length, tone and formatting rules
• Use reusable AI Context information
• Create prompts for missing tasks
• Test changes with a real request
• Restore a built-in prompt through Reset to default

T3AF keeps built-in prompts in the connected extension while storing project-specific custom prompts separately. This lets teams adapt AI instructions without changing the extension’s original source code.

Inside the 122 Prompt Templates and Four Extensions

T3AF currently provides 63 editable prompts across nine categories and more than seven connected extensions. Administrators can review these instructions from one central catalogue instead of searching through separate extension settings or source files.

The prompts available in a project depend on the installed prompt-enabled extensions. Each connected extension contributes its own categories, while AI Foundation stores the shared templates and makes them available at runtime.

Prompt Area	Typical Tasks	What Teams Can Control
SEO	Meta titles and descriptions	Length, keywords and search intent
Content	Writing and rewriting	Tone, structure and reading level
Translation	Localised content	Locale, formality and terminology
Media	Alt text and descriptions	Detail, accessibility and length
Chat and search	Responses and queries	Voice, relevance and fallback rules
Development and shared tasks	Technical workflows	Format, safeguards and conventions

Why Categories Matter

Categories help teams find the right instruction and assign ownership. SEO specialists can manage metadata prompts, localisation teams can review language rules, and developers can maintain technical instructions.

This structure also makes duplicated prompts, conflicting rules and missing safeguards easier to identify as more AI features are installed.

Prompt or AI Context? Know Which One to Change

Poor output does not always mean the task prompt is wrong. The AI may simply be missing the brand or audience information needed to complete the task correctly.

AI Context explains who the organisation is. A prompt explains what the AI must do.

Use AI Context For	Use Prompts For
Brand voice and audience	Output length and format
Company information	Number of results
Preferred terminology	Task-specific restrictions
Language style	Translation behaviour
Reusable content rules	Missing-information handling

Use AI Context when information should influence several AI workflows. Use a prompt when the rule applies to one task.

For example, a formal brand voice belongs in AI Context, while a 150-character limit belongs in the meta-description prompt. T3AF documentation describes AI Context as reusable brand information and prompts as task instructions.

Connecting Both Layers

Prompts can use reusable context variables:

Write one meta description for the current page.

Use this brand voice:
{brand_voice}

Follow these content rules:
{content_rules}

The task remains reusable, while the selected AI Context profile supplies the relevant brand rules.

How to Troubleshoot Poor TYPO3 AI Output

Begin with the visible problem rather than rewriting the entire prompt.

Output Problem	What to Inspect
Metadata is too long	Character limit
Copy sounds generic	AI Context and tone rules
German is too informal	Locale and “Sie” instructions
Output contains hype	Forbidden language
Claims are invented	Factual boundaries and fallback rules

Follow a Five-Step Diagnosis

1. Identify the AI feature producing the result.
2. Open AI Foundation → AI Prompts.
3. Filter by the relevant extension or category.
4. Review the prompt, variables and active AI Context.
5. Change one instruction and test again with the same input.

Keep the source content, provider and model unchanged during the first test. This makes it easier to see whether the prompt edit caused the improvement.

If results become worse, use Reset to default, change one variable and test again. This controlled approach is recommended in the T3AF documentation.

Prompt troubleshooting works best as a controlled test, not a complete rewrite after one weak result.

Step by Step: Creating or Customising a Meta Description Prompt in T3AF

The process starts by checking whether the connected extension already provides a suitable prompt. There is no need to create a duplicate when an existing instruction can be adapted.

Step 1: Open the AI Prompts Module

In the TYPO3 backend, go to:

AI Foundation → AI Prompts

The overview displays prompt categories contributed by installed, prompt-enabled extensions. AI Foundation provides the central module, while connected extensions supply the prompts used by their features.

Step 2: Filter by Extension

Use the extension filter to select the extension responsible for the task.

For a meta-description workflow, choose the extension that provides the relevant SEO or content-generation feature. Review the available prompts under that extension before creating anything new.

This helps avoid:

• Duplicate prompt instructions
• Similar prompts with conflicting rules
• Custom prompts assigned to the wrong extension
• Unnecessary project-specific overrides

Step 3: Review the Existing Prompts

Check whether an existing prompt already matches the task.

When a suitable meta-description prompt exists, open it and review its title, prompt type, variables and instruction text. You can then customise the existing instruction according to the project’s SEO and brand requirements.

When no suitable prompt exists, select Create New Prompt.

Step 4: Add the Prompt Details

Complete the fields shown in the prompt form:

Category
Choose the extension category where the prompt should be available.

Title
Use a clear name that editors can recognise, such as:

 

B2B Meta Description

 

Description
Briefly explain when the prompt should be used:

 

Generates one concise meta description for B2B service and product pages.

 

Prompt type
Select or enter the prompt type required by the connected feature. Prompt types should remain stable because extensions use them to locate the correct custom instruction at runtime.

Prompt instruction
Add the instruction that should control the output:

Write one meta description for the supplied page content.

Requirements:

• Maximum 150 characters, including spaces
• Write for B2B decision-makers
• Use the primary keyword naturally when relevant
• Match {brand_voice}
• Follow {content_rules}
• Use clear and specific language
• Do not use hype, emojis or quotation marks
• Do not invent claims or benefits
• Return only the meta description

Keep any variables required by the connected extension. Removing a required placeholder may prevent the prompt from working as expected.

Step 5: Save the Prompt

Review the selected extension, category and prompt type, then save the record.

AI Foundation stores editor-created custom prompts centrally. The connected extension can then select and load the saved prompt when its AI feature runs. The documented resolution process can use an explicitly supplied prompt, a saved custom prompt or the extension’s built-in default.

Step 6: Confirm It Works

Run the relevant feature on one real page and confirm that it uses the new or customised prompt.

Check that:

• The correct prompt is selected
• Required variables are populated
• The result follows the requested length and format
• The AI does not introduce unsupported information

The workflow is therefore simple: open AI Prompts, filter by extension, review what already exists and create a new prompt only when the required instruction is missing.

Seven Prompt Improvements That Produce Better TYPO3 AI Output

Strong prompts replace subjective requests with instructions that editors can test.

Improvement	Weak Instruction	Better Instruction
Define the tone	Make it professional.	Use direct sentences, technical clarity and restrained language.
Add negative rules	Make it suitable for our brand.	Do not use emojis, hype, invented statistics or competitor names.
Specify the format	Write some title ideas.	Return exactly five titles without explanations.
Set measurable limits	Keep the summary short.	Write two sentences with a maximum of 45 words.
Define the locale	Translate into German.	Use German for Germany, formal “Sie”, and preserve product names in English.
Handle missing context	Answer using the page.	If the source is insufficient, do not guess. Return “Insufficient source information.”
Add a final check	Return the answer.	Check the length, format and unsupported claims, then return only the final answer.

T3AF recommends defining the required length, format, language and tone, as well as content the AI should avoid.

Use negative instructions for problems the team has actually observed. A long blacklist can create conflicting rules and make the prompt difficult to maintain.

Locale also needs more detail than a language name. Teams may need to define regional terminology, formality, date and number formats, or product names that must remain untranslated.

A final self-check can improve consistency, but it does not prove factual accuracy. Editors should still review legal, technical and commercially sensitive content before publication.

Managing Prompts Across Multisite and Agency Setups

Agencies may manage several sites in one TYPO3 installation, multiple brands with different language rules, or completely separate client projects. Each setup needs a slightly different governance process.

Multiple Sites in One TYPO3 Installation

AI Foundation stores shared prompt templates that connected extensions can load at runtime. This reduces duplicated instructions and gives administrators one place to review prompt changes.

A multisite team can use one approved meta-description prompt to define:

• Maximum length
• Required output format
• Keyword treatment
• Rules against unsupported claims

Because a shared prompt can affect every editor using the connected feature, changes should be reviewed before release.

Different Brands Need Different AI Context

A shared task prompt does not require every client to sound the same.

The prompt can define how to create a meta description, while each AI Context profile supplies the relevant audience, brand voice, SEO terms and language style. T3AF supports reusable brand profiles that connected extensions inject into prompts.

An agency can therefore keep one SEO framework while changing the brand information applied to each site.

Separate TYPO3 Installations

T3AF documents shared prompt storage and project-level customisation, but it does not describe automatic one-click synchronisation of edited prompts between independent installations. Agencies should use a controlled release workflow instead.

A practical process is to:

1. Maintain an approved master prompt catalogue.
2. Store versions in documentation or source control.
3. deploy approved defaults through extension configuration.
4. Apply client-specific changes in the relevant installation.
5. Retest after extension, provider or model updates.

This creates reusable standards without hiding project-specific decisions.

Create an Agency Prompt Standard

Use names that show the task, locale and version:

 

SEO_META_DESCRIPTION_V2

CONTENT_SERVICE_INTRO_V1

TRANSLATION_DE_FORMAL_V3

CHAT_SUPPORT_FALLBACK_V1

 

Assign a clear owner to each prompt area:

Prompt Area	Suggested Owner
SEO	SEO lead
Content	Content strategist
Translation	Localisation lead
Compliance	Legal or compliance reviewer
Technical workflows	TYPO3 developer
Publishing approval	Client or editorial lead

Clear naming and ownership prevent prompt changes from becoming anonymous backend edits.

A Safe Prompt Editing Workflow for B2B Teams

Prompt editing should follow a controlled release process. The four stages below describe governance steps rather than four required technical environments.

1. Draft

Change one instruction at a time and record the reason for the edit. Preserve the previous approved version before making a substantial change.

2. Test

Run the updated prompt against a small but representative test set:

• A normal page
• A short or incomplete source
• A technical page
• A promotional page
• A multilingual page
• Content containing risky or unsupported claims

Keep the source, provider and model unchanged during the first comparison so the effect of the prompt remains visible.

3. Review

Ask the designated prompt owner to score the output.

Criterion	Score
Correct format	/5
Brand alignment	/5
Factual accuracy	/5
Length compliance	/5
Editorial usefulness	/5

A practical approval threshold is 20 out of 25, with factual accuracy scoring at least 4 out of 5. Any invented or unsupported claim should trigger an automatic failure, regardless of the total score.

4. Release

Apply the approved prompt to the production workflow and record:

• The release date
• The prompt owner and approver
• The previous and current versions
• The pages and languages tested
• The intended output behaviour
• The rollback procedure

AI Foundation’s documentation warns that prompt changes affect all users of the connected feature, so large production changes should be coordinated with the relevant permissions and editorial owners.

Keep the Reset Option in Mind

T3AF provides a Reset to default option when an edited prompt performs worse than the built-in version. The documentation recommends restoring the default, changing one variable and testing again.

Resetting makes experimentation safer, but it is not a replacement for version control. Teams should preserve important approved prompts outside the editing interface so they can compare changes, restore project standards and explain why an override was introduced.

Common TYPO3 AI Prompt Customisation Mistakes

Adding Too Many Rules at Once

Large rewrites make testing harder. Rules may conflict, and teams cannot identify which change improved or weakened the result.

Change one instruction at a time whenever possible.

Putting Brand Information Into Every Prompt

Brand voice, company details and general terminology belong in AI Context. Prompts should focus on the task, format and output requirements.

Repeating brand information inside every prompt creates duplication and makes updates harder.

Using Subjective Instructions

Weak:

• Make it attractive.
• Make it engaging.
• Make it premium.

Better:

• Use short paragraphs.
• Lead with the customer problem.
• Avoid promotional adjectives.
• Include one concrete benefit.

Observable rules produce more consistent results than vague creative labels.

Forgetting the Output Format

A useful answer can still be unusable if it returns commentary, Markdown or five options when the TYPO3 field expects one clean value.

Specify exactly what should be returned.

Optimising for One Test Page

A prompt that works for a detailed service page may fail on:

• Thin pages
• Product listings
• Translations
• Technical documentation
• Pages without a clear keyword

Test with varied content before releasing the prompt.

Treating Prompts as a One-Time Setup

Review prompts when:

• Brand guidelines change
• New extensions are installed
• AI providers or models change
• Editors repeatedly rewrite generated output
• SEO requirements change
• New languages are introduced

TYPO3 AI Prompt Customisation Checklist

Before saving a prompt, check:

• Is the task clearly defined?
• Is the intended audience specified?
• Is the required output format explicit?
• Are length limits measurable?
• Does it use the correct AI Context placeholders?
• Are forbidden words or unsupported claims addressed?
• Does it explain what to do when information is missing?
• Is the language, locale and formality clear?
• Has it been tested on more than one page?
• Is the prompt owner or approver recorded?
• Can the previous version be recovered?
• Does the output require human review before publication?

Prompt Transparency Turns AI From a Toy Into Infrastructure

AI output quality begins with the instruction. When prompts remain hidden, teams can only regenerate weak results and hope for an improvement. Visible prompts let them identify the problem, adjust the rule and test the result.

A central prompt library also makes quality repeatable. AI Context defines the brand, while prompts control the task. Agencies can therefore reuse tested instructions without forcing every client to sound the same.

As more editors and AI extensions are added, prompt governance becomes essential.

The real advantage is not that T3AF gives TYPO3 more prompts. It gives TYPO3 teams ownership of the instructions shaping every AI result.

Conclusion

Better AI output begins with better instructions. By making prompts visible and editable, T3AF helps TYPO3 teams replace repeated regeneration with a more controlled process for improving quality, brand consistency and factual accuracy.

Teams can explore the capabilities on the AI Foundation for TYPO3, follow the configuration and development guidance in the T3AF documentation, or review the extension’s code and releases in the T3AF GitHub repository.

The value of T3AF is not simply having more prompts. It is giving TYPO3 teams ownership of the instructions shaping every AI result.