Tools Used In This Recipe
If your startup founding team can build your product, sell it, and develop marketing copy to help convince potential customers of its value, consider yourself lucky. Of those three skills, the ability to develop marketing copy is usually the lest common. But new AI Large Language Models (LLMs) can offload the burden of generating copy for you, if you know how to use them. OpenAI’s GPT-3 and companion interface ChatGPT have amazed users with their ability to generate high-quality content for a wide variety of requests. Tools like Jasper AI provide context around LLMs, turning GPT-3’s blank canvas into a series of tools to address common use cases in marketing and more.
In this Recipe you’ll use both GPT-3 and Jasper AI to address a very common startup challenge - developing Search Engine Optimization (SEO) tags for your website.
Before you begin
In this Recipe we’ll be generating meta title and description tags for Startup Recipes itself, to show you how to use the tool. You can apply these techniques to your own startup or business to generate meta tags for your website, but you can also start from the Recipe on identifying search keywords for your startup. As a best practice you should build one page on your website for every search term or phrase you want to rank for. You can follow the processes in this Recipe in this case as well; you will just generate one set of meta tags for each keyword or phrase.
To get the best results from this Recipe you should have a 3-5 sentence description of your product or company already written (you can use Jasper AI to help with this too, or use the Recipe on Brand Positioning Statement). You should also have a sense of your brand voice (but when in doubt, stick to ‘professional’).
In this Recipe we’re going to use two tools to accomplish the same objective, to show you the process. Jasper is the more expensive tool but it includes a lot of helpful context in its user interface, so it’s definitely a better choice for people who are new to Large Language Models.
Go to the Jasper home page and click Start For Free.
Once you have set up your account you’ll be directed to the Jasper home page. All new accounts start with a 5 day free trial with 10,000 credits (about 7,500 words). Jasper also uses a clever onboarding process where you can earn additional credits by trying out its key features.
Once you have gone through as much of the onboarding flow as you choose, click the Templates button to get to the full list of tools.
Jasper has dozens of workflow functions designed to help users accomplish specific tasks. It’s worth clicking through the categories at the top to see all of the options Jasper has. Chances are you’ve needed help with a few of them in the past.
In this Recipe we’re looking to generate SEO tags for our site, so click the SEO button at the top of the page.
For this Recipe we’re generating SEO tags for our website’s homepage, but if your startup has multiple products or services you can use the same process to generate SEO tags for those using the other options listed here. To make this Recipe relevant to as many startups as possible we’re going to use Jasper’s Chat feature instead of its structured walkthrough tool, so instead of clicking any of these options click the Chat option on the left.
As the link implies Jasper’s Chat tool uses an interactive messaging interface. Instead of filling out a form with multiple fields to generate your output text you can type a request using natural language. To ask Jasper to generate SEO tags for Startup Recipes we use the following request:
Write three creative SEO title and meta description for a product called Startup Recipes. Title tags should be under 60 characters, and meta description tags should be 160 characters or fewer. Use a professional but informal brand voice. The product description is: Startup Recipes is a curated, actionable set of how-to guides for time-crunched entrepreneurs. Validate your startup idea, launch landing pages, do competitor research, build a pitch deck, raise a VC round, manage social media, run paid ads, optimize SEO, make your first hire, and more. All using free or low-cost tools and best practices from high-growth startups across the globe.
Large Language Models perform best when you provide specific context, so let’s take a look at this request. First off we ask for several candidate responses (in this case three). Because these will be generated by an AI there will be subtle differences between them, and this will give us enough options to pick one that works better for us. We also used the adjective ‘creative’. This tells the AI to increase the variations between the options.
In the second sentence we tell the AI to limit each tag to a specific maximum length (we chose those limits because those are the maximum character limits for Google search results for the title and description fields. In the third sentence we tell the AI what brand voice to use, and lastly we provide the description of the product.
Once you’ve entered the text for your prompt you can hit Enter or the Send button. Your prompt will show in a chat interface similar to text message applications, and Jasper’s AI will begin working on your request. Most requests get a reply within a few seconds but it’s possible for Jasper and other LLM tools to take a few minutes to generate a response if they are backed up or the request is complex.
Jasper will generate its response in an interface designed to look like a chat reply. If Jasper used one of its templated tools it will list that at the top.
Just like we asked, Jasper generated three candidate title and description combinations for the prompt we used. At the bottom of this (and every) response are a few buttons. The eye icon will hide the response. Since Jasper is using a chat interface it ‘remembers’ the responses it gives so that you can give it new prompts based on its output. For example you could type “translate these to French” and Jasper will understand that when you say “these” you mean the response it just gave you.
The other buttons include a thumbs up & down option (you can rate Jasper’s output as helpful or unhelpful), regenerate (which tells Jasper to create a new response based on your previous prompt), copy (to your clipboard so you can use the output in other applications), or to open the output in a document (Jasper includes a writing tool similar to Google docs).
OpenAI’s GPT (Generative Pre-trained Transformer) is an LLM that provides a more bare-bones interface, but is equally as capable of generating quality output as Jasper and is significantly cheaper since it only uses a pay-as-you go model (the most expensive LLM it offers costs $.02 per 750 words generated). We’ll use the same prompt to show you how GPT-3 compares to Jasper.
Start by visiting the OpenAI home page and signing up for an account.
The GPT-3 playground has an interface that at first glance looks like Jasper’s chat, but instead you type commands directly in the main interface and GPT-3’s responses show up in green directly below your entry.
The right side of the screen has a few options to allow you to affect the output. Most of them you can ignore but there are two that you will want to evaluate for your use case.
The Model dropdown allows you to change the LLM that GPT-3 will use. OpenAI has four models each with different levels of data, and the amount of data used directly influences the quality of the result. Unless you are very cost-conscious it’s fine to leave this set to the default, davinci.
The other setting you need to adjust is Maximum Length. LLMs work on a token basis, and they use the number of tokens in the prompt (the request you type in) combined with the amount of text in the proposed output as part of the algorithm. The maximum that GPT-3 can support is 4,000 tokens and there’s no penalty for using that setting other than potentially having the model take a few extra seconds to think, so leave this at 4,000.
With those settings in place we’ll enter the same prompt we used with Jasper.
In a few seconds it will return a set of results.
As with Jasper you can regenerate the results as much as you wish until you find a result that you like. For this prompt we asked these LLMs to generate both a title and description tag so in this case we can actually mix and match - we can pick a favorite title tag and a different description tag.
Updating our website
Depending on what tool you use to host your website there will be different places for you to set the title and description for your pages. As an example we will be using the Startup Recipes landing page that’s built on Carrd (this Recipe show you how to implement the landing page if you are interested). Check with the platform documentation for your website if you use a different tool (or ask the person who built it for you if you don’t know how to update this content yourself).
To update the title and description for our landing page on Carrd, log in and from the dashboard select the landing page we have already built from the dashboard. Click the pencil icon to launch the editor.
In Carrd since there is only one page you set the title and description when you publish. Click the disk icon to start the publish process, and enter the title and description copy you chose from the LLM output.
Once you see the Carrd success modal, your new copy is live.
Now we’ll use the Chrome browser tool to verify that our new copy is in place. Click View Site and your landing page will open in a new browser. Once it does, right-click anywhere on the page and click Inspect (if you are using Firefox, Safari, or a different browser the menu name will be different, but they all support this type of functionality). A window will open on the bottom of the screen with several tabs. This window hosts a number of developer tools built into the Chrome browser. Click Elements, then click the arrow next to
<head>…</head> to expand the page’s meta tags.
A new set of HTML tags will open, and you can scroll through these until you fine the one marked
<title> and one marked
<meta name=”description>. Make sure these match what you entered.
It may take search engines a few days or even weeks to crawl your site and pick up the new copy, but once they do you’ll start seeing the benefit of the new SEO-optimized tags.
In this Recipe we used two different LLMs to generate SEO tags for our website, but these tools have limitless potential to streamline text and image generation for a variety of challenges startups face. If you haven’t already explored them you should prioritize getting familiar with at least one of these tools, because the impact on your business will be significant.
The contents of this Recipe are © Innovation Works, Inc. and are licensed under CC-BY-SA 4.0 . Contact us with questions or feedback, or to learn more about our structured program in Entrepreneurism based on Startup Recipes.