5 AI hacks that SEO editors need to know

The tasks that AI can take on within an editorial team are diverse. Checking texts, writing summaries, producing translations and much more. However, the lack of factual accuracy of the systems often remains a problem in text creation. To counter this, a more innovative approach was required. Where artificial creativity can be used to great effect with the greatest possible freedom, it can possibly make editorial work more efficient and simpler within the narrow confines of smaller task areas. This approach has proven to be absolutely right and has helped us to develop various prompts for everyday tasks. These are used exclusively in projects in which the client has explicitly commissioned the use of AI technology and are presented below.

Tool-based work is part of everyday life in content marketing: determining search volumes, identifying semantic keywords, analysing user behaviour and much more is part of the toolkit of every good SEO editorial team. The AI working group has derived instructions for action from various test series in order to enable all colleagues to use AI offerings in a targeted and effective manner in everyday editorial work. We will outline some of these possible applications below in order to sharpen the focus on alternative AI applications in editorial offices.

Delivering error-free texts is one of many basic requirements for successful collaboration with paying customers. If errors are repeated, what would otherwise be the best job is soon history. That's why diva-e applies the 4-, or even better, 6-eyes principle. In addition to the editor who is responsible for creating the text, at least one other human person checks the text in the proofreading department. AI services can be used as a supplement here to recognise and correct gross errors as the first checking instance. This not only increases quality, it also makes downstream professional editing easier and can save costs.

In practice, it is clear that the AI needs to be curbed in its creativity to achieve better results. After all, it should recognise misspelled letters and missing commas and correct grammatical errors, but not rearrange or delete entire sentences or create additional content. Furthermore, the corrected result should be output in full and the corrections made displayed. This makes it possible to copy and paste the corrected text and also allows the editor to check and evaluate the result.

At diva-e, we use the ChatGPT systems from expansion stage 3.5 and Bard for this purpose. Bard is currently the preferred tool for this purpose due to its greater reliability. While the associated prompt can be customised, the following instruction has proven itself for standard cases:

"I would like you to check a German text for correct spelling, punctuation and sentence structure. Then print out the corrected text in full and briefly list the changes you have made. Wait until I have posted the text to be corrected."

Shorter versions have also led to good results. However, reliability fluctuates at peak times because either no corrections are made, corrections that have been made are not listed, or errors are certified even though misspellings occur.

Reliability: 85 %

Recommended AI: Bard

Meta data, title and description for the output in the SERPs are rather hard work in the otherwise varied work of an SEO editor. This is repetitive content that has to be created according to a fixed pattern for changing keywords. The structure and brand are the same. What is still easy to do with a few "metas" quickly becomes a tedious chore with dozens or even hundreds of pages.

Instead of torturing student workers, this task can be performed by artificial intelligence without any grumbling. And even if the quality fluctuates, AI does not always follow the instructions exactly and therefore an optimal result is not always achieved, the creation of meta titles and meta descriptions by AI is a real relief, especially in the masses. If hundreds of keywords need to be written, meta data that is 80 per cent optimised and produced at the touch of a button is always better than no meta data at all.

From version 3.5, ChatGPT has proven to be quite reliable in the production of metadata. Caution: The quality fluctuates. Particularly at peak times - when the computers in the USA are booted up on working days - optimum results are often no longer achieved. Tests with up to 50 keywords have worked well, although not all prompt specifications have always been executed exactly.

Here are the prompts:

Production of meta titles

"I need meta titles for various keywords of the brand <brand>. In the title, the keyword must be at the beginning, the brand after a | at the end. Remember that the title can be a maximum of 55 characters long: Please design the titles according to the following example: <example>.
I need three title suggestions for each of the following keywords: <keyword1>, <keyword2>, <keyword3>, ... Please cluster the result for each keyword in a table."

Similarly, AI can design suitable meta descriptions:

"I need meta descriptions for various keywords of the brand <brand>. These should contain the main keyword, list up to three USPs and end with a CtA. The aim is to increase the click probability. The user should obtain information on the product page and can also buy directly there. Please remember that the description must not be longer than 155 characters. Use the following example as a guide for the descriptions: <example>

I need three description suggestions for each of the following keywords: <keyword1>, <keyword2>, <keyword3>, ... Output the result in tabular form sorted by keywords."

Reliability: 75 per cent

Recommended AI: ChatGPT version 3.5 or higher

Like the production of meta data, the identification of user questions, known as W questions in technical jargon, and the resulting production of FAQ content is a standard procedure and a recurring task in everyday SEO editorial work.

Answering the right W-questions increases the likelihood that users interested in the topic will find their way to the page. There are numerous specialised tools for determining the right W-questions. It is striking that the various SEO applications for determining W-questions provide a large number of very detailed questions. AI remains on the surface in this task and mainly reflects more general questions about the what, the how and the why. Therefore, the use of specialised tools is always recommended at this point, which allow an even deeper insight into the user's world of thought.

Where AI sometimes lacks depth, Bard in particular more than makes up for this with its full-service approach. Unlike pure W-question tools, ChatGPT and Bard can also provide the right answers straight away. This means that an initial draft for an FAQ module can be completely filled with text within a few minutes. Our recommendation in this use case is clearly in favour of Bard. ChatGPT has no access to data analyses and therefore derives questions and answers purely on the basis of word order probability.

As a Google application, Bard is a decisive step ahead here. When asked about the methodology used to determine the most important questions, Bard replies: "In this case, data from search engine log files was used." The system therefore already has access to Google data and can generate added value from this.

The prompt for determining the W questions is then very simple:

"List me the <number> of most frequently asked user questions for the keyword <topic> for a search engine optimised landing page."

The second step involves the AI-based production of an initial FAQ draft:

"Please answer the <number> of questions listed. The aim is to satisfy the user's need for information. The answers should be a maximum of <number> characters long."

Limiting the number of characters should help to keep AI creativity within limits. In many cases, the result is already quite close to what is required in the customer project. However, facts need to be checked and, as always, the draft should be the starting point for editing.

Reliability: close to 100 per cent

Recommended AI: Bard

SEO editors don't always start a customer project with a blank sheet of paper. Often, content already exists, but it either does not follow a clear structure, is incoherent in terms of content or has simply not been adapted to the target group in terms of style and tone. Instead of disposing of this content and starting from scratch, it can make sense to utilise what already exists. AI tools can help to revise large volumes of text in a short space of time. For example, if the style needs to be standardised, modernised or the user approach needs to be more direct. For example, texts that argue too passively can be formulated more actively using AI.

The type of revision is defined by the specific requirements in the customer project. Therefore, there can be no generalised prompt in this usage scenario. Rather, the creativity of the editors is also required for prompting.

Considerations that must precede an AI revision:

• How should content be revised?

• What output do I want to receive?

• What information does the AI need to deliver optimised content?

Example prompt from the revision of an old customer text - topic and content are derived from the original text:
"Please revise an existing website text. Redesign the content so that it reads as a personal recommendation and active help for the user. The author speaks from his own experience as an expert. The text should sound like a personal recommendation that provides the reader with detailed information on the topic. The reader is addressed directly and is on first-name terms. The writing style should be relaxed and aimed at a young target group. Anglicisms are permitted. Add a personal tip from the author at the end of each paragraph."

Reliability: 85 per cent

Recommended AI: ChatGPT version 3.5 or higher

In the editorial environment, the complete production of a text with AI can probably be considered the supreme discipline. In previous use cases, artificial intelligence was forced into a creative corset with existing content or within clear instructions that set clear limits to the tendency to hallucinate. In text production, however, we cannot avoid letting ChatGPT, Bard or JasperAI off the leash. To date, however, all AI offerings have had difficulties in the realisation of complex copywriter briefings, which are the norm in everyday editorial work. In numerous trials, we have not even succeeded in having customer specifications fully implemented by an AI. The aim can therefore only be to use artificial intelligence to create an initial draft text, which is then refined by an editor and adapted to the customer's exact requirements.

This procedure is not only highly recommended for quality assurance. Comprehensive editorial revision also ensures that the requirements of German copyright law are met and that legally compliant texts can be delivered. As a rule of thumb: for a text to be protected by copyright, at least 50 per cent + 1 word must be written by a human pen.

But how do I get the AI to write a basic text that the editorial team can then continue to work with? In practice, it turns out that the systems need context in order to understand the task and objective.

The following information should therefore be included in the prompt:

• Who is the sender?

• Who is the recipient?

• Which topic is being addressed?

• What is the objective?

• Are there any special features of the content?

• Information on style/tonality

• Length and structure of the draft

Example prompt:

"You are a <specification sender>. Write a website text on the topic <topic> for the target group <specification recipient>. Firstly, address the typical risks <specification content features>. The text should list the benefits <specification content features> and answer the most important questions about the topic. The aim is to encourage the user to <objective>. The reader is addressed directly, he is addressed as a salutation. The text should be between 2,500 and 3,000 characters long".

As a result, we received texts on various topics and objectives that provided a sufficiently good basis for further processing. Of course, this cannot be the be-all and end-all. In general, it can be said that the more context the AI is given, the better the result to a certain extent. However, this also has its limits, as too much complexity is no longer implemented correctly. Generally valid statements can therefore not be made. In a specific application, only trial and error really makes sense.

Reliability: 45 per cent

Recommended AI: ChatGPT version 3.5 or higher

Used correctly, artificial intelligence is a valuable support in everyday editorial work. diva-e is already successfully using AI in the use cases shown here and several others. Further use cases are constantly being tested in the editorial department. Development in this area is rapid and it is not yet possible to say what the systems will be capable of in a few years' time. Where reliability is not yet 100 per cent at the moment, this could be the case tomorrow. What is certain is that AI will continue to push its own boundaries and take on new tasks in the future. We will accompany this process and utilise the new possibilities efficiently for our customers.

Then subscribe to our newsletter now and don't miss any updates!