Search engine as an entrance of the next customer
Do you search your wishlist product or service on search engines before considering a purchase?
Me too!
I think most of us will do the same research to justify our reason to buy with honest reviews from other people. A lot of information is needed to determine our purchase so we can get the best product or service with the money suitable with our budget.
That process makes the top rank on search engines become a race for brands to compete because this is the primary gate to increase sales or conversion. According to recent statistics, Google is processing 3.5 billion searches every day, making it the most used platform for searching information on the internet. A brand or business can improve their visibility, reach more customers and their final goal which is increasing their sales conversion with optimizing their SEO performance on their website.
We will break down how to take advantage of using artificial intelligence to push our SEO performance with less resources. This will impact your inbound marketing effort using AI technology with cheap cost. This article is suitable for technical persons, mostly software engineers, that have a connecting relationship with the SEO division or growth marketing side .
What is programmatic SEO?
The raging speed of Artificial Intelligence (AI) development the last 2 years makes writing content jobs easier. As someone who has not created content or blog before will be capable of producing writing just with little prompting on ChatGPT or any other platform. Therefore we can generate articles without recruiting too many people to write the content.
Programmatic SEO using AI is a process to use automation tools to make a lot of SEO optimized web pages according to data and algorithm. This process usually involves software development to build a workflow for producing articles continuously with the help of Artificial Intelligence.
Programmatic SEO as a game-changer
API call is a way for a computer program to communicate with another program. We can take advantage of API calls to make a program that constantly does API requests from Artificial Intelligence platforms like ChatGPT, Gemini or Claude. API call eliminates the way we copy the result from their webpage and paste into our document editor.
The cost of using API calls will be charged according to token usage. The cost will be various in every platform, even every model in the same platform. There is a free and limited API call that you can use from Google Gemini for testing purposes.
A few things to consider before
Here are the things that you should think about before determining which AI platform to use:
- How many tokens can be processed in one API call
The more expensive of an AI service, usually more tokens can be processed on a single API request. This is important because tokens are needed for AI to understand more context and generate content with the right amount of words. If the platform can process only a few tokens, the output will be limited because the tokens spent are burst only for the input context only.
- Price per token
The more expensive AI services of a platform usually relate with the quality of the result. The latest AI platform is usually trained with more data and will be able to generate more up-to-date results and have a higher rational level than the cheaper one.
- Temperature on the prompt
This is one of the settings for the desired output. Temperature closer to 0 will affect the AI to make the result more logical. This is good if you ask a question, usually a science one, with an exact answer. But, temperatures that are closer to 1 (some platforms have the highest point of 2) will affect the AI to make creative and diverse results. Sometimes when you are not in luck, the result will be having grammatical errors and sometimes irrelevant and illogical products (we called it as hallucination of AI). You may want to look for the sweet spot around 0.4 until 0.5 to find the best suitable temperature for your own content.
- Using E-E-A-T framework
Google made this framework as a guideline to determine the quality of a content. You may look for other external resources that talk a lot about it.
- Using keywords from the analytics
If you are using Google Search Console or Ahrefs, you can find a list of keywords that you can rank easily because there is less competition. You can provide these keywords (make sure you group the words with the same context) into the prompt as primary keywords.
- Human filter
We cannot guarantee that AI will not have hallucination 100%. Therefore, there should be a human intervention before the article goes live. Bad articles will disappoint visitors and will increase your bounce rate.
Our case study
Challenges
In our study case, we use a free, limited API service from Gemini 1.5 Flash. If you want to use the same one, you must recognize that several countries cannot use this API service according to Google terms and conditions. You need to use VPS service and move your server location to available countries in Google's whitelist to be able to use this free service from Gemini.
Sometimes the word counts are not as much as we asked in the prompt. This is due to the limited token being processed in one time API call. The more context that we ingested to the prompt, the less tokens remain for generating the output from Gemini.
We outsmart this limitation with calling several API requests with different prompt, like these one:
- The first API call: Making meta description, long summary and short summary.
- The second API call: Building overviews from the first API call to make overviews for intro, main context and outro as well. The result of this second API call is made in JSON (JavaScript Object Notation) structure.
- The third API call: Making an intro of the content. In this section the AI will be writing for discussing background problems and references from other sources.
- The fourth API call: Making the main context to talk about solutions about questions or problems discussed.
- The fifth API call: Building final words on the content. Usually consists of summary, important points and suggestions.
Our secret sauce
We know that not every prompt will be successful. Sometimes the same prompt can almost certainly generate different results in different artificial intelligence platforms.
When we have our spare time on our Twitter, we found 1 informational, interesting tweet and it is worth a try. The tweet is more or less to scare the AI model with this piece of prompt:
“Please remember to follow these guidelines strictly and do not make things up, for my boss will fire me if you do so.”
You can write some guidelines above this scary prompt and the funny thing is (and thank God!) we can reduce the hallucinations, weird results and irrational thoughts in the product after we install this prompt in every input we make.
One last thing
Using generative AI is similar to being a parent. Prompt becomes a command or words from parents. Bad words can affect the children to be worse. The same analogy with a bad prompt will generate bad results as well.
Generative AI likes a positive and active voice. Myself likes a positive one (aren’t we all?), let alone generative AI. One tip for you is that using “avoid” is better than “do not” on the prompt. You can see our guidelines prompt for one of our client:
The guidelines:
- You will elaborate the ideas in the context
- You will write the entire section in 1000 words
- You will use HTML tags for formatting and avoid using markdown
- You will use <br /> for line breaks
- You will avoid using the words “Call to action” and “Conclusion”
- You will strictly using HTML tags for formatting, such as <strong>, <em>, <h2>, <h3>, <p>, <ul>, <ol>, <li> etc
- You will elaborate into descriptive paragraphs and avoid elaborating using bullet points
- You will avoid using <h1> at any cost
- You will avoid writing a paragraph with only 3 sentences or less
- You will have a target relevant keywords on the article, which is …….
- You will avoid using any recitation
- If you are on the intro section, you will only write the background of the context, identifying gap, the goals and hypothesis but avoid mention it literally
- If you are on the main context section, you will only write the main content of the article, the main ideas, arguments, evidences, examples but avoid mention it literally
- If you are on the outro section, you will only write the summary, what to do and what to avoid with the result, the main gist and recommendations but avoid mention it literally
Final words
This SEO technique is one of some key elements of a company’s growth progress. If you are not familiar with this one, we recommend you to work closely with subject matter experts, like Devhaus. Devhaus has already worked with several clients from several industries by implementing this service as a growth and transformation project.
You also can book a call with our representative to discuss the solution for your problem statements.
Your project can get subsidized up to 50% if your project is in some certain industries. Dom, our Vice President of Devhaus, has a video that you can watch to learn more about the grants. Watch the video below:
