Can AI write manuals for you?
Artificial Intelligence is a useful tool. But can AI be used to write operations or product manuals? If so, how good are the results? What are the issues with AI?
Artificial Intelligence is developing fast. It is increasingly useful but not perfect … yet.
Read on to discover how current versions of AI can and cannot be used to help write manuals, issues to be aware of, and how we recommend using AI today.
How does AI work? (A quick introduction).
Two types of AI are useful when writing manuals:
Large Language Models (LLMs).
AI image generation.
Examples include ChatGPT/OpenAI, Gemini, Claude and Canva.
A Large Language Model (LLM) is a type of artificial intelligence designed to understand, process, and generate human-like text. Powered by deep learning, these models are trained on massive datasets—including books, articles, and websites—to learn the statistical patterns and structural relationships of language. Rather than thinking or comprehending like a human, an LLM operates through sophisticated mathematical prediction.
AI generates images by using neural networks to interpret text prompts and reverse engineer them into visual content. Instead of "drawing" from scratch, the AI starts with a canvas of random static (noise) and calculates probabilities to gradually sculpt and refine the noise into a coherent, high-definition image. Before creating anything, the AI is trained on vast datasets containing billions of images and their corresponding text descriptions.
Can AI write a manual from scratch?
No. Not really. Not yet. But it is still useful.
Prompting AI to “write me a manual for a restaurant” or “write instructions for a microwave” will NOT generate a comprehensive, cohesive, comprehensible, illustrated manual that is useful or on-brand for your operation, or for the procedures or products you wish to document.
AI is designed to respond to the specific prompts it is given and to give concise answers based on content that already exists.
Asking AI for a restaurant manual gives a one page reply. Even modifying the prompt to be more specific - e.g. “write me a comprehensive 50 page manual for a burger restaurant in the UK” - still only generates 5-pages ‘borrowed’ from the manual of a leading burger brand (so also a copyright infringement).
Similarly, asking AI to “write me a comprehensive user manual for a microwave oven sold in the UK” gives 2-3 pages of generic instructions that won’t apply to the specific model you offer.
Like any data engine, AI suffers from the old adage; ‘garbage in, garbage out‘.
To generate potentially useful copy (text), AI must be given very specific and very well considered prompts, one subject at a time. AND the content you need must already exist, somewhere on the internet.
So, instead of asking AI to write a manual, it is better to ask for specific content, such as “give a comprehensive standard operating procedure for receiving ingredients from delivery to storage in a fast food burger kitchen with individual under-counter fridges and freezers, a walk in cold room, and racked ambient storage, that is in line with UK food safety standards, temperature records and regulations that would earn a 5 out of 5 food hygiene rating from the EHA - in step by step bullet format”.
Such a prompt requires that you already have a good understanding of the topic, but might generate a fairly useful response from AI.
BUT BEWARE, most responses from AI require additional research, verification and re-writing before they can be effectively (or legally) compiled and used in a manual (see further below).
None of these prompts would generate images or illustrations. You can ask AI to include images in a text, but this usually results in AI giving links to existing copyrighted images or websites, as examples.
AI is good at generating graphics and photos one-by-one, if fed with example images and a good prompt (but is less good at step-by-step line drawings for SOPs, machinery instructions or product manuals). Image generation for commercial purposes (so for both operations and product manuals) incurs a fee or subscription and the images generated must still be added to your document manually.
In summary, AI can not (yet) design or compile a complete, accurate, fully illustrated and on-brand manual in your chosen software (e.g. Microsoft Word or Adobe Illustrator) at the touch of a button. It can generate components, but you will need to compile and format them manually.
So this begs the question - Is the amount of expert knowledge, time and effort (i.e. cost) needed to give AI effective prompts, understand what AI gets wrong, edit responses, compile answers into a manual and generate AI images one by one, not just the same as as writing the manual yourself (or getting Manual Writers to do it all for you)?
The answer is probably a balanced one:
If they were mutually exclusive, you or an expert (like Manual Writers) can develop better manuals than AI at the same or less cost.
However, using AI to SUPPORT your manual writing, (i.e., as a research tool and to shortcut some images), can save time and money.
So, how can AI support manual writing?
Rather than expecting AI to do it all for you, use AI’s strengths to help you write a better manual.
AI is good at:
Research - Give AI the right (specific) prompt and let it do the leg work to scour the internet for answers and resources. AI can reduce days of desk research and web-surfing into seconds.
But beware, AI responses are biased and limited by the way you ask the question and not always accurate or cohesive.
Always check, correct and adapt responses to fit current local regulations and your specific operation or product.
Notes and organisation - Use AI to take and compile notes, transcribe and summarise meetings, structure and improve outlines, manage data gathering, cross-reference information and organise assets to be used in your manuals.
Grammar and quality of writing - AI does a good job of fixing, clarifying and reframing text.
Give it texts you want to improve, or ask it to generate foundational texts when you have ‘writer’s block’.
Be careful to tell it what style of writing you need and double check the results.
Be aware that AI sometimes mixes up grammar, spelling and naming conventions between UK English and American English.
Proofing - Ask AI to ‘read’ a text and give suggestions that would make instructions clearer or add considerations you may have forgotten.
Incorporate any suggestions that are useful.
Image generation - Use AI (at a cost) to generate images and graphics to illustrate your manual. Note:
AI images tend to be generic, so are best used to convey an idea, rather than to illustrate your specific processes or products.
Be careful to avoid ‘AI Slop’ (obviously fake and generic images) in marketing materials (e.g. the cover page of a product instruction manual). AI may sometimes use copyrighted images as a base, leading to copyright infringement - always check that commercial licenses for the images you use are watertight. (Some AI services, such as Adobe Firefly, claim to only generate images that are safe for commercial use).
Basic formatting - use the AI built into the software you use (e.g. Copilot in Word or Firefly and Sensei in Adobe Illustrator) to create templates for page formatting and document layouts.
Be aware that these may sometimes be generic, so your competitors may end up with similar formats.
What are the issues with AI?
The two biggest issues are copyright infringement and the generic nature of responses.
Copyright infringement:
You must NEVER use the intellectual property of other individuals or companies in any commercial manual, without their written permission or license. That stands whether the manual is only used internally (e.g. an operations manual) or is distributed to customers (e.g. a product manual).
AI (mainly) uses multiple sources from the internet to generate text or images. Most responses thus include at least some elements belonging to other individuals or companies.
Be very careful incorporating AI responses or images in your manuals. Always check permissions and licenses, if possible. At the very least, use responses only as a starting point. Develop and modify text and images so they are in your own words and style and are specifically adapted to your brand and ‘voice.
Students using AI to cheat on essays might just get a ticking-off. You might be sued for large amounts of money or face other litigation!
Generic information and images:
AI doesn’t know you, your company, your operation or your products. It can only generate text and images only from what already exists on the internet. Information and images generated may even be from your competition, inferior and out of date practices or worse products.
This means that information and images generated by AI typically do not:
Reflect the specifics of your processes or products.
Consider any innovations, new practices, commercial secrets, proprietary designs etc. that set you apart from the competition.
Include the way that you want things done. (Or the specific way your product works).
Consider the make-up, preferences, culture, mission, vision, style and ‘voice’ of your team and your brand.
Include all possible up-to-date considerations, information, legislation, regulations, safety issues or technical processes specific to your industry, operation or products.
Other issues with AI:
You should also consider the following when using AI to support manual writing:
AI is biased by the prompts it is given: Asking AI the same question in different ways can generate different responses. AI uses your prompts to formulate its answers in the ways it thinks you want it to. So try to keep prompts specific and neutral.
E.g. asking AI “What are the minimum safety notices needed in a manual to prevent prosecution?” or “What are the safety notices needed in a manual to ensure the safest possible operation” will give very different responses.
In this example, a better neutral prompt might be “Detail safety notices required for safe operation, in line with current best practice and regulations in the UK”.
AI combines multiple sources: AI quite often mixes several sources in its responses. This means you might be given information that combines practices and regulations from completely different industries, countries and legislations. AI may also cut out information that would have been useful to you, had it stuck with one source, because it wants to add other sources. Ask for responses that are specific to your industry and location and request information from specific relevant regulatory bodies or sources.
AI has an American bias: Be careful to ask for responses in your language. AI often spells everything in American English and uses American naming conventions.
Pleonasm / Redundancy/ Stuffing: AI has a habit of padding out lists or using more words than necessary to express an idea. A single word or concept is supported by additional, unnecessary words that mean the same thing (e.g., "this thing, and that thing" when both just mean "stuff"). Check any text you use scans well, sounds natural and isn’t just ‘padding’.
AI ‘hallucinates’ steps: AI sometimes adds steps to processes and procedures that are unnecessary or not relevant to your operation or products. All procedures must be checked by a knowledgeable person in your organisation - removing steps that are not needed.
Be careful copy-pasting! Most AI responses end with a disclaimer such as “AI responses may include mistakes’, include hyperlinks for other brands, or use images that are obviously AI generated (E.g. AI has a habit of decorating lists with colourful icons). Ensure these are removed when copying responses directly into manuals.
