Transcription
What if I told you that trading perplexity in search GBT like Google is the biggest mistake when it comes to AI searching? Most people still use it like a Google search, but you can actually get much better results by using just one single query. The difference is all about how you ask.
In this video, I'll share six search prompting methods you can use on top of your existing questions so you can fully utilize the potential of these smart AI search engines every time. Let's go!
In this video, I'll be switching between perplexity and search GBT so you can see how this method works on both platforms. For perplexity, I'm using the PR search with the default model set to GBT 4.0 model and ChatGPT Plus for accessing the SE GBT features. Next, I've turned off the memory and custom instruction setting to make sure the results are unfiltered.
The first one is what I call the debate analysis method. Basically, it forces the AI search engines to go beyond surface-level analysis and add different contrasting views, so you will get a more complete picture of a certain topic to inspire your own thinking. It works particularly well when you need to understand multiple perspectives on some emerging trends or controversial topics.
First, we have the base questions, such as "What are the trends in a certain topic?" and then in the same PR, ask it to analyze key findings or trends in this topic. For each one, require it to have contrasting views, and this is the most important.
For example, let's say I'm searching for the latest trend in the topic of responsible AI. For now, you can see it gives me the key trends happening under this topic, like governance and regulations, ethical framework, and addressing bias, which is good and with the source. But now, if we add this to the prompt and ask it to include the contrasting view for each point, you will see the response will have much more depth.
Although the points are still similar, you can see now it adds the key developments for each one and also the multiple viewpoints. For ethical framework, the debate is all around how to balance ethical considerations with business objectives. So you can quickly get a sense of where the tension is going on for each one.
It works great, particularly if you're not familiar with a topic, and so you understand both sides rather than just getting factual information or assuming you will get them automatically from the AI search engines.
This method not only works for trend-related questions; you can apply it to any type of question, so you can always get contrasting views for each key finding, and it definitely makes a big difference.
Probably you're using AI search engines as part of your content creation process to do research and get insights. I want to share a valuable resource from HubSpot about creating trustworthy content. It's a playbook about navigating E-A-T in the AI era. I put it in the description for you to download for free.
This playbook reviews the strategies that help HubSpot achieve remarkable results in search performance by creating experience-rich authoritative content that meets Google E-A-T standards while also sharing practical tips for the execution of these strategies.
I highlight their narrow and deep strategies, which is the approach for smaller teams to build content authority without extensive resources. It's practical and speaks directly to the common resource challenges that most businesses face today, showing how to achieve better results by focusing deeply on just a few core topics where you have real expertise.
I recommend you download this in the description below for free, and thank you, HubSpot, for sponsoring this video.
All right, the next is the blind spot method. Whenever we're searching online, we're looking for solutions or finding information to make better decisions. This is why AI search engines are so helpful because they help you analyze information faster and identify potential risks.
This method aims to identify factors that might not be immediately obvious to you but could be important for your better decision-making. First, we present the decision question, such as "Should I start a YouTube channel?" and then we quest for long-term impact analysis, like three years, five years, or even ten years.
Then, we look for overlooked hidden factors. For example, I have a decision question: "Is it good to move to a new country at 30?" Without using this method, you will see the response is still good. Perplexity will automatically pull different sources like YouTube, Reddit, and mention both the benefits and challenges, like career opportunities, having a fresh start, and challenges like culture adjustment needing time to build up a network.
They are still helpful responses, but if we use this prompt framework and request it to consider the impact in five years and hidden factors, we are forcing it to think from a long-term perspective.
So the response, you can see, focuses more on the five-year impact. It now mentions personal growth and enrichment. Even for building a social network, it now frames it as an upside rather than a downside, as from the previous prompt.
Also, the positive impact for families that have kids because of the long-term impact. What's more amazing is the hidden factors, which we won't get in the first PR, such as the healthcare system and visa requirements. These are real challenges that people may overlook.
So just having this small tweak to your PR will give you a much more comprehensive view and help you to look beyond just the short-term impact for better decision-making. So definitely try this method for whatever decision question you have and which you need to have some factual information to back up.
Next, which is one of my favorites, is the trend tracking method. Unlike getting the contrasting view for a trend or topic, it focuses on comparing data from different time periods. This summarizes how a certain topic evolves over time, and you can identify which trend is just hype and which is really an ongoing trend.
It avoids outdated information, so it works best for topics that change fast. First, define your research topics and time frame. I would suggest using a broader topic and not too niche, and then compare the current trend with the time frame you specify, like six months ago or one year ago, and finally summarize the future predictions for you.
For example, cybersecurity is definitely a topic that is evolving fast, and I want to quickly understand the key developments in this topic. So in a normal search, I just ask about the latest developments in cybersecurity. You can see it will give you the core areas to explore, like AI-driven threats, ransomware, and cloud security.
They are still good as they give you ideas on key areas of a certain trending topic. But if we use this prompting method and force it to compare how cybersecurity evolves or changes over time, like a year or six months, you will see the response will be so much better.
It starts with previous trends, like the rise in ransomware, hybrid work challenges, and until recently, the rise in AI in cybersecurity, to the latest future predictions like AI-driven attacks and quantum computing, which is definitely on trend now.
It even gives you a summary table, and that's why I like this method. Because instead of just getting some generic topics or areas, it helps me to understand the topic more completely, identify which are actually the most recent trends, or which is just noise, so I can always focus in the right direction.
This method works particularly well for research topics that are evolving fast, like technology trends, market developments, and industry changes. I like this method because the results will always be better, as it clearly shows me the progressions of the key events and helps me to spot the changes easily.
The next one is learning path creation. As growth marketers, we need to constantly upskill ourselves as technologies change so fast. Perhaps it's technical skills, soft skills, or academic subjects, but the information is just too overwhelming.
So this method shows you clear progressions and gives you directions on whatever the skills or resources are needed for learning purposes. First, specify your learning goal, perhaps learning UX skills, and then request it to be broken down by different learning levels and also to include relevant skills, tools, courses, and certifications.
The last step is optional; sometimes the AI search engines will just automatically include that for you in the search results. For example, let's say I'm learning UX. In a normal search, I may just ask how to learn these skills. It will give you a basic approach, like understanding fundamentals, enrolling in courses, and learning practical skills.
I would say these are not bad, but still, the path is not clear enough if I am an absolute beginner. So now, when we ask it to build a learning path for the particular skills and group it by beginner, intermediate, and advanced, you can see the results will be much more detailed.
For each level, it has a clear learning objective, like beginner level learning fundamental principles, learning wireframing, and to the intermediate level to enhance proficiency, learning advanced user research, and to the advanced level about design leadership with links to courses and certifications.
This way, it's like having a clear roadmap instead of just random directions to achieve your learning goals. Of course, you can even ask it to give you a reasonable timeline, and it works for any type of skills, even language, business, or creative skills.
The next is using search operators, and this is super useful. You may have used search operators on Google; there are special commands that help to refine the search results to make them more precise. But you can actually use them on AI search engines as well, and this makes them even more powerful when combined with the AI search capabilities.
Basically, for most of the search operators, they work well on the AI search engines, and in particular, the three search operators are used the most. First is the file type operator to search for particular types of files, such as PDF files.
Another one is the site operator to search for results from a particular website, such as some authoritative websites or domains ending with .edu, .gov, or .org, which are usually primary sources for academic research and official reports with bias.
The last type is date range-related, after and before, so you can filter the results within a particular time frame. You can also combine them with the Boolean search operators like AND, OR, NOT, and they all work well on perplexity and SE GBT.
For example, I'm searching for a technology trend outlook, and using the contrasting view prompting method we just talked about, we can now add the file type operator to just filter the PDF files. Then you will see it will pull resources of PDF files, like this one, the report from McKinsey, and all the PDF file sources.
The best thing is it will present the search results as we discussed, with a summary of key findings and also contrasting views. Another example is I'm deciding whether to apply a work policy for my company, and then using the blind spot method, we can add the site operator to just filter the sources from domains ending with .edu, .gov, or .org to make sure the sources are less biased.
Now you can see the sources are all coming from institutions like the US Bureau of Labor Statistics and this detailed report from the European Trade Union Institute, and Stanford report. So the search results will be even better now because you're requesting it to pull from particular authoritative sources and present the results with the prompting method that we discussed.
This will make a huge difference. If you know other authoritative sources for your industry, definitely use them so you can greatly improve the search results quality.
Another way is to use the date range. Let's say I'm learning everything about machine learning, and so using the learning path creation method, I can then add after and before with the year and month. In this case, it's last year and until this November.
This way, I can always make sure the results are the most recent ones. You can see the sources are all within the date range I provided. Sometimes you may find the date is out of range from what you requested simply because it always references the first publish date but not the last update date, so keep this in mind.
For now, I find this works better on perplexity than search GBT because of its more diverse source of information than search GBT. To take a step further, you can even use formatting options to make sure AI presents you the search results in your desired formats, such as a table, review matrix format, or even a structured report format. Just be creative here.
The next one is the fact-checking method. You know one of the most common challenges when using AI search engines is hallucination. I'm going to say every AI model will hallucinate, but just to what extent? There are ways to minimize that in our prompts, and this method will force AI to find supporting claims and primary sources for verification and not just simply interpret the facts from what it understands.
First, you present your claims, such as "90% of startups fail in the first year," and then request AI to find you the primary sources and statistical evidence. For example, I have this claim: "Our brains are adaptable during periods of uncertainty," and then I will add this to the prompt and ask it to find the direct quotes from primary sources and to list statistics with dates.
Of course, you don't need to copy the exact wordings; just use your own wording, and it's fine, but you get the idea. Now you will see it will pull direct quotes from sources that have mentioned this claim. For this case, there is a source from MIT about how the brain deals with uncertainty, and that is an authoritative source.
It also finds statistics with a date, so you can make sure this claim has evidence to support it, and if the date is recent, which is so important. For this claim, you can see it says it's likely true given what it finds.
Now, there is another method which is even more simple and is one of my favorites for any type of claims. You just add this to the prompt: "Is it true?" and then it will immediately force AI to find the supporting evidence from sources for you.
Like in this case, I'm verifying if "80% of chat users never touch the powerful features for productivity." You will see it will find some sources about chat usage, though it says the exact figures are lacking, and this statement is plausible without enough support.
So using this method, you will quickly understand how likely a statement or claim is true or not. Of course, you still need to fact-check yourself, and it is a must whenever you do online searches, but this prompt will definitely save you so much time. Personally, I use that a lot, and it also works perfectly on perplexity, so make sure you use it.
All right, these are the prompting methods you can start using right away, and you realize it's all about how we think and approach a problem when we're using the AI search engines. You don't need to copy the exact wordings in these prompts; just understand the logic behind them and customize your own.
There are actually two more powerful prompting frameworks I personally use for better decision-making, and I share these along with detailed examples in my community. You will also get all the prompt templates I share on my channel to create your own version. If you're interested, check out the link in the description to join.
And be sure to check out this video about the powerful AI prompting hacks if you haven't done so yet. I'll see you next time. [Music]