AI in practice: content and website maintenance for SMEs

The marketing manager of a medium-sized business proudly presents a new blog: twelve posts in four weeks, all generated by AI. One look is enough. Every article begins with “In today's digital world” and ends with “Conclusion: it depends”. Not one contains a figure, an example or a statement with any bite. Traffic is barely measurable.

In brief: AI writes quickly, but knows nothing about your own business. The benefit comes from a process in which people provide the facts and retain responsibility. This article shows where AI really saves time in content and website maintenance, where it becomes risky, what a reliable workflow looks like and which legal issues matter, supported by figures on adoption among SMEs.

Send an enquiry

41 per cent use AI, while half struggle with skills

The figures are clear: around 41 per cent of businesses use AI productively, and roughly a third plan to do so. The most common applications are text and translation, used by just over seven in ten adopters, marketing and communication by around half, customer service by just over four in ten and data analysis by just under a third. A clear majority of users report an improved competitive position.

The obstacles are equally revealing. More than half cite a lack of AI skills within the team as the biggest barrier, while a third report higher costs than expected. The findings come from a representative survey of around 600 businesses with at least 20 employees. In practice, the deciding factor is whether somebody in the organisation knows what the tool is suitable for and what it is not.

Where AI helps and where it becomes risky

The greatest benefit arises where a person subsequently checks the result and takes responsibility:

• Rough drafts and outlines: structure, initial wording and alternative headings as a starting point, not a finished product.
• Rewriting and shortening: adapting existing copy for other channels and making specialist material easier to understand.
• Metadata: titles and descriptions at the required length, followed by a content check.
• Translations as a basis for final review by someone with subject expertise.
• Images for blogs and social media where a real photograph is not required, with close attention to style and brand fit.
• Recurring customer-service answers as suggestions for an employee, not automatic replies.

AI is unsuitable for unchecked statements that must be correct: figures, legal information, prices and product details. Language models sound convincing even when they are wrong.

Website maintenance: underestimated time drains

In everyday maintenance of a business website, AI mainly saves time on routine tasks: suggesting alternative text for images, generating meta descriptions for many pages, checking old copy for clarity, deriving FAQs from existing support enquiries and adapting content for social media.

AI also helps technically, with troubleshooting templates, explaining unfamiliar code and writing small data-cleanup scripts. What it does not provide is responsibility for the result. An automatically generated alt text that misdescribes an image can harm accessibility more than it helps, so final checks remain a human task.

A website-maintenance example shows the difference. Forty images are missing alternative text. The naive approach is “AI, write alt texts for me”. The result is usually unusable because the model does not know the page context. The useful approach supplies that context:

Role: You are an editor for the business website of a
plumbing business in Braunschweig.

Task: Write alternative text for the image.

Context:
- Page: "Bathroom renovation"
- Image content: installer fitting a level-access shower
- Purpose: shows workmanship, not a product photograph

Rules:
- maximum 125 characters, no "image of", no keyword stuffing
- describe what is visible, not what it is supposed to mean
- neutral, no marketing

Output: only the alternative text.

The difference is the brief. Without context, the output is “An image of a bathroom”. With context, it becomes “Installer fitting a level-access shower in a renovated bathroom”. The first text harms accessibility; the second helps. A person must still check it, because no model knows whether the image really shows a shower.

Brief, draft, fact-check, refine

A simple four-step workflow has proved effective. First, briefing: provide the audience, tone, key messages and facts rather than having AI invent them. Second, drafting: AI supplies structure and rough copy. Third, fact-checking: verify every figure, quotation and legal statement against a reliable source. Fourth, refinement: add voice, examples and practical relevance. This is where interchangeable copy becomes valuable content.

The order matters. Checking facts only at the end means correcting copy built on false assumptions. Starting without a brief produces generic content, recognisable by sentences that say everything and nothing.

In practice: ten articles, two errors and a rethink

A typical scenario showing how such a case unfolds in practice.

A supplier with 60 employees begins using AI in marketing. In the first attempt, an intern generates ten blog posts in a week. The copy sounds polished but includes an incorrect delivery time and an invented certification. Sales notices after a customer asks about it.

The second attempt looks different. There is a three-part rule: facts come from the business, AI supplies the wording and a person approves it. Specifically, the sales manager provides five bullet points and verified figures for each topic. AI builds the structure and draft. The marketing manager checks each figure against the internal source and writes the introduction and examples herself.

After a quarter, four articles are produced instead of ten generic ones, and people actually read them because they contain things only that business knows. The time saving is not “a finished article in five minutes”, but faster intermediate steps: outlining, rewriting, metadata and English translation.

Another finding often surprises people: the bottleneck is never the AI, but the willingness to provide it with real information.

Data protection, copyright and the AI Act

Three subjects belong on the agenda. Data protection: customer data and internal documents should not be fed unfiltered into public AI services. For business use, data-processing agreements and plans that do not train on submitted data are sensible. Copyright: the legal position of AI-generated images is not settled in every respect. Care is needed when imitating recognisable creators' styles or protected brands. Transparency and regulation: the European AI Act introduces phased obligations, including content labelling and employee competence.

A short internal policy is usually enough for everyday use: which tools are allowed, which data may be entered, who checks the result and how content is labelled. This provides clarity without slowing adoption. This article does not replace legal advice.

Does Google penalise AI copy? That is the wrong question

Search engines assess whether content is helpful rather than whether a person or machine wrote it. Mass-produced, superficial material loses because it contributes nothing new. Conversely, carefully edited, fact-checked articles with clear structure benefit.

The same applies even more strongly to visibility in AI answers. Clear, evidenced and well-structured material is cited. Original data, concrete examples and real practical experience are the part no AI can supply, and precisely the part that determines visibility.

Start small, but commit

The best starting point is one specific use case with measurable benefits, such as meta descriptions for the twenty most important pages or drafts for a monthly blog post. Assign a responsible person, choose a tool and write a short policy. After four to six weeks, review what saved time and what created more work.

Experience quickly shows that the benefit lies less in producing copy at the push of a button than in accelerating intermediate steps. That is where expansion pays off, and where the value can be demonstrated when management asks.

Conclusion: the difference is in the process

For SMEs, AI is no longer an experiment but one tool among many, with clear strengths in routine work and rough drafts and clear limits around facts and responsibility. Establishing briefing, fact-checking and refinement as fixed steps saves time without sacrificing quality. Skipping review will eventually publish mistakes more costly than the hour saved. The difference lies in the process.

Frequently asked questions

Short answers to common questions about AI for content and website maintenance.