How to create a new Taxonomy Topic
This document is a overview of topic creation best practices and how to create valuable, targeted, high intent topics in the Delivr Taxonomy.
Topic Creation Best Practices:
What does a TOPIC do?
Delivr.ai measures the INTENT a person has for a given subject. When creating a topic, you are finding people researching, viewing and otherwise engaging with your subject of interest. There are different kinds of intent. I might be just browsing, I might be looking to buy right now. It’s important to keep in mind what you are trying to measure when creating a topic.
From a technology standpoint, we are leveraging the foundation of most LLM’s, the concept of “similarity” between different words, phrases or intents. We are measuring the similarity between the intent, and buyer behavior.
Things that make a good topic and description
SEO keywords ← the same thing that makes your ads work, or your website rank. What are your customers searching for on the internet?
LSI Keywords ← closely related keywords/concepts to your core SEO keywords. Ex. “Weight loss” might include “gym”
Being focused and concise
The role of your business context
It’s important to understand that what you are trying to accomplish is important, but shouldn’t be included in the actual topic or description. The focus of a topic is on finding what your AUDIENCE is researching and cases about, not what you care about.
That being said, your business context can change how you think about the intent you are trying to track. If you are launching a new product, you might want to focus on what you think your differentiation would be, or the broader market pain.
Get in your customers heads
What is your audience looking for on the internet? We’ll go find it! This is a very similar problem to buying adwords on google. When people search, they are showing intent based on the keywords they use. Delivr will look at the context of the page, url, domain to see things that are related to your key topics, but you have to decide what your audience thinks is interesting.
Poorly structured topics create irrelevant audiences.
Intent Gotchas
What are the kinds of things that don’t really work for intent?
There has to be something to find, or it will find ANYTHING instead.
When tracking intent, there’s a natural inclination to get really really specific. “Hey, let me see how many people are interested in my thing!” If your topic is not well known, is very new, or doesn’t have lots of relevant information on the web about it, the chances are you’ll be scooping up traffic that isn’t closely related to what you really care about.
Quick test:
Google search your topic. If you don’t see 5+ pages of relevant search results, change your topic name until you do! If you see some relevant results, and some irrelevant results, add keywords until the results are more relevant.
Don’t include anything about the audience you are trying to reach
Intent is focused on what your AUDIENCE is trying to find on the internet. If you include details about your audience, you will be tracking people that are interested in your AUDIENCE, not your service.
Quick Test:
Does your topic or description include any details about the people who might buy your service? Use your product? Remove that and focus only on the product or service.
Don’t include anything about your business model, or how you charge, or whether you are luxury or low cost, or other “meta” descriptions
In marketing copy there’s an inclination to include information about what the people you are looking to reach care about. “Interested in luxury goods”, “focused on quality testing” because those characteristics represent your brand or service. For intent, those kinds of descriptions will just find activity of people interested in those subjects in general.
So a description of “health care product XYZ intended for users interested in sourcing good materials and quality control” will go find activity of people researching quality control! Anything included in the description is part of the intent.
If you wouldn't buy adwords for it, maybe don’t include it!
Quick Test:
Are you describing the product or service, or interests of people who like the product or service? Google search the adjectives you are using, do you get relevant results? If no, remove them.
Too many words.
The model is looking at the words you are using to measure intent. Using too many words will confuse it, by giving it too many relevant things to track. Providing a list of competitor names won’t be effective, as there’s not enough context to ensure those brand names are specific to what you care about. Being focused on a singular subject is where it’s at.
Quick Test:
Is your description just a couple sentences? Could you share it with your mother and have her understand it? Is it focused on just a single subject, or does it include un-related details to that subject?
Don’t include details about why the intent includes certain keywords
Ex. “By focusing on these terms, the taxonomy segments audiences who are actively evaluating providers and financing options—rather than those only seeking general information.”
The intent algorithms can’t “target” specific audiences in the way you think. It’s going to measure the intent based on the keywords provided directly, and ANY information included will be part of the intent result. So in this example.. “Seeking general information” is now part of the intent!! This is the complete opposite of the intended result, so don’t include anything you don’t directly want to measure intent of.
Intent can’t target a geography or “near you”
Intent is looking for contextual interest across the internet. It doesn’t know where you are, or where your business is located. It’s not a replacement for local SEO. Instead you can gather information about anyone in North America interested in a service, and then create an audience that targets your ICP.
Intent can’t measure quantities / sizes / or thresholds
Intent isn’t logically analyzing quantities, or thresholds. So you can’t build intent like "People researching soda cans over 12 oz” or include something like “people under 40”. The model doesn’t understand the concept of “larger than” or “smaller than” and doesn’t handle quantized results well.
Case Study
Lets do a case study of a topic and pick apart what kinds of things it will track for intent -
GOAL: Find consumers that are searching or in the market for a gutter solution for their home.
TOPIC: ROOF GUTTERS FOR MIDDLE AGED HOMEOWNERS THAT DON’T WANT TO CLIMB LADDERS
DESCRIPTION: A HIGH PERFORMANCE GUTTER SOLUTION WITH LOW MAINTENANCE FROM A TRUSTWORTHY CONTRACTING COMPANY WITH HIGH RATINGS AND LOW PRICES. OUR PERFECT BUYER IS MIDDLE AGED HOMEOWNER WHO IS AFRAID OF FALLING AND OF LADDERS
Picking things apart
First.. Lets look at the Topic Name:
We are selling “ROOF GUTTERS” .. but the topic also includes “MIDDLE AGED HOMEOWNERS” and “CLIMB” and “LADDERS”.
This means the topic will track intent to ALL of those keywords and the intent behind them, and even if someone is researching roof gutters all day long, those pages likely don’t include “MIDDLE AGED HOMEOWNERS” or “CLIMB LADDERS” and so it won’t be a good intent score. The additional terms “confuse” the model into having poor matches for everything.
Worse, you might be getting intent based on articles about middle aged homeowners or services that sell to that audience (because they use those keywords!)
A Better Outcome
So let’s clean this up. Here’s some better topics that are more specific to the INTENT we are trying to measure.
NAME1 : Gutter Guard & Leaf Filter Installation
NAME2 : Clog-Free Gutter Solutions
NAME3 : Professional Gutter Replacement Services
DESCRIPTION: This intent represents seeking professional installation or replacement of low-maintenance, clog-free rain gutter systems. It involves finding contractors and comparing quotes for solutions like gutter guards and leaf filters that provide lasting protection.
You can see the topic is much more specific to the service people should be searching for and researching (showing they are in market), and our description includes things related to roof gutters that will help us scoop up more intent about our main intent subject “ROOF GUTTERS”
Intent Type Examples
Service Intent Example
Here’s a “more real” example and how we can get to a good topic and description.
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent topic, and description.
Topic:
Description:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic.
Recommend a two sentence description of what someone researching the service would be looking for using related LSI keywords and SEO best practices specific to the service.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt in Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent topic, and description.
Topic: People Exploring Legal Support for Social Security Disability
Category: Business Services
SubCategory: Legal Services
Description: People Exploring Legal Support for Social Security Disability are individuals beginning the SSDI application process or anticipating obstacles—such as understanding eligibility, gathering documentation, or completing government forms. This audience may include those recently diagnosed with chronic conditions or experiencing new work limitations, along with caregivers helping them navigate the process. They’re not always facing a denial yet but want reassurance that their application is complete and accurate. The leverage point is legal guidance: disability attorneys help reduce delays, avoid errors, and improve approval odds. Most begin seeking help after reading eligibility rules or encountering early confusion with paperwork.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic.
Recommend a two sentence description of the service using related LSI keywords and SEO best practices specific to the service.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 10” or “under 65”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : Legal Assistance for SSDI Applications
NAME2 : Social Security Disability Lawyers for Initial Claims
NAME3 : SSDI Eligibility & Legal Guidance
DESCRIPTION: Represents seeking legal guidance to understand Social Security Disability (SSD) eligibility rules and application requirements. This intent involves finding a disability attorney to assist with initial claims and paperwork to improve the chances of securing benefits.
Brand Intent Example
The goal of brand intent is to identify an audience that is researching a brand, or markets, products, and services closely related to that brand. This can be tricky as brands are often common keywords.
Ex. “Toast”
A crunchy cooked piece of bread
A restaurant point of sale platform ← this is the brand
The process of toasting something
Since intent is semantically based, it’s important to ensure we provide enough context to the model to separate out the different meanings of the same word.
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent brand, and description.
Brand:
Brand Description:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that brand compared to other common uses of the words in the brand name.
Recommend a two sentence description of the intent using related LSI keywords and SEO best practices uniquely identify that brand.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt in Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent brand, and description.
Brand: Toast
Brand Description: Toast POS offers comprehensive restaurant solutions with online ordering, contactless payments, inventory management, and real-time reporting. Its cloud-based system streamlines operations, improves guest experience, and fosters growth through integrated hardware and software.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that brand compared to other common uses of the words in the brand name
Recommend a two sentence description of the intent using related LSI keywords and SEO best practices that uniquely identify that brand. Be concise and specific. Do not include details about the audience.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : Toast Restaurant POS System
NAME2 : Toast Point of Sale Solutions
NAME3 : Toast Restaurant Management Software
DESCRIPTION: This intent identifies engagement with integrated point-of-sale (POS) systems and comprehensive restaurant management platforms. It encompasses interest in solutions for online ordering, contactless payment processing, inventory control, and kitchen display systems (KDS) for the food service industry.
Event Intent Example
Events come with built-in audiences, these kinds of intent topics track the interest in the event, and build an audience of attendees and those researching the event itself. The goal here is to identify topics and keywords specific to the event itself. You don’t want to include anything that could be confused with other events, or even what the event is about (because that will open your intent topic to an audience of people interested in the subject, but not the event).
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an event you’d like to track intent for, and a description.
Event:
Event Description:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that event compared to other common uses of the words in the event name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices that are specific to the event.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt In Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an event you’d like to track intent for, and a description.
Event: Dreamforce
Event Description: The audience for Dreamforce primarily includes Salesforce customers, partners, developers, and industry professionals interested in cloud computing and customer relationship management (CRM). Attendees are often looking to network, learn about new Salesforce products and features, and gain insights into industry trends and best practices from thought leaders.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that event compared to other common uses of the words in the event name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices that are specific to the event.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : Dreamforce: Salesforce CRM & AI Conference
NAME2 : Dreamforce Salesforce Event
NAME3 : Salesforce Trailblazer Conference (Dreamforce)
DESCRIPTION: This intent identifies engagement with content about Dreamforce, the annual Salesforce conference for its global community of Trailblazers. It includes research into the agenda, session tracks for Sales Cloud and Service Cloud, AppExchange partner solutions, and major updates to the Einstein AI platform.
Product Intent Example
Product intent allows you to
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent product, and description.
Product:
Product Description:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that product compared to other common uses of the words in the product name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices for that product.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt In Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent product, and description.
Product: Apple Safari
Product Description: The audience for Apple Safari includes Apple device users such as Mac, iPhone, and iPad owners who are looking for a web browser that prioritizes speed, privacy, and user-friendly features. This demographic typically consists of tech-savvy individuals, students, professionals, and anyone invested in the Apple ecosystem seeking to optimize their online experience.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that product compared to other common uses of the words in the product name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices specific to the product. Be concise and specific. Do not include details about the audience.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : Apple Safari Web Browser
NAME2 : Safari Browser for macOS and iOS
NAME3 : Apple Safari Secure Privacy Browser
DESCRIPTION: This intent represents interest in Apple's native web browser, Safari, for macOS, iOS, and iPadOS devices. It captures engagement with topics like Intelligent Tracking Prevention, Passkeys, browser extensions, and performance benchmarks on the WebKit engine.
Solution Intent Example
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent solution, and description.
Solution:
Solution Description:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that solution compared to other common uses of the words in the solution name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices specific to the solution.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt In Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent solution, and description.
Solution: Outsourced Customer Service
Solution Description: The audience for outsourced customer service includes business owners and managers seeking to enhance their customer support without the overhead of hiring in-house staff. This topic appeals to small to medium-sized enterprises looking to improve efficiency and customer satisfaction, as well as larger companies aiming to reduce costs or streamline operations. It may also interest service providers that specialize in outsourcing solutions.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using urls and domain traffic. Include enough keywords in the name to uniquely identify that solution compared to other common uses of the words in the solution name. Recommend a two sentence description of the intent using related LSI keywords and SEO best practices specific to the solution. B
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : Business Process Outsourcing (BPO) for Customer Support
NAME2 : Outsourced Call Center and Contact Center Services
NAME3 : Omnichannel Outsourced Customer Experience (CX) Solutions
DESCRIPTION: This intent represents research into third-party providers for handling customer-facing communications and support operations. It includes engagement with contact center as a service (CCaaS), BPO pricing models, and strategies for improving customer satisfaction (CSAT) through scalable, 24/7 support.
Function Intent Example
Example Prompt:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent of a technical concept or function, and a description of the desired user intent. Technical Concept/Function:
Description of Intent:
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using URLs and domain traffic. Include enough keywords in the name to specify a particular intent related to this concept, distinguishing it from other contexts or applications of the same function (e.g., distinguishing "SSL certificate providers" from "SSL implementation tutorials"). Recommend a two-sentence description of the intent using related LSI keywords and SEO best practices specific to this technical concept.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
Example Prompt In Use:
You are an expert marketer creating an audience using intent to target prospects.
Here’s an intent of a technical concept or function, and a description of the desired user intent. Technical Concept/Function: JSON Web Token
Description of Intent: The audience for the topic of JSON Web Token (JWT) primarily includes software developers, particularly those involved in web development and API security. This may also extend to system architects and IT security professionals who are implementing secure authentication processes in applications and services. Additionally, businesses looking to enhance their web applications' security would also be interested in understanding how JWT works.
Use the related information to recommend three improved names that will be used by a large language classification model to analyze intent using URLs and domain traffic. Include enough keywords in the name to specify a particular intent related to this concept, distinguishing it from other contexts or applications of the same function (e.g., distinguishing "SSL certificate providers" from "SSL implementation tutorials"). Recommend a two-sentence description of the intent using related LSI keywords and SEO best practices specific to this technical concept.
Rules:
Be concise and specific.
Do not include details about the audience.
Don’t reference the conference locations or dates.
Don’t include URL’s in the recommended names or descriptions.
Don’t include concepts like “near me” or “local”
Don’t include quantities, sizes, thresholds like “larger” or “smaller” or “over 50”
Return Data in the following format:
NAME1 : RECOMMENDED NAME
NAME2 : RECOMMENDED NAME
NAME3 : RECOMMENDED NAME
DESCRIPTION: RECOMMENDED DESCRIPTION
RESULT FROM GEMINI 2.5 PRO
NAME1 : JWT Implementation and Coding Libraries
NAME2 : JSON Web Token (JWT) Security Vulnerabilities
NAME3 : Token-Based Authentication using JWTs
DESCRIPTION: This intent represents interest in the use of JSON Web Tokens (JWT) for stateless, token-based authentication and secure information exchange in APIs and web applications. It encompasses the implementation of signing algorithms like HMAC and RSA, the management of security claims and payloads, and the mitigation of common token vulnerabilities.
Delivr Topic Review Process
Delivr has a quality control process for topics submitted through our process. The best way to have your topic approved quickly is to use the example prompts we’ve provided above, as that will create something that matches the format we expect.
If these best practices are not followed, we will:
Reject the topic (if there’s not enough information to complete it correctly, or if the topic requested won’t be effective)
Edit the topic name to follow our best practices
Edit the topic description to follow our best practices or improve it’s effectiveness
This review process typically takes 2 days, but can take longer depending on the amount of edits required for the requested topic. Topics that don’t require edits (following our best practices) can be approved quickly.