A short video has been doing the rounds in which a creator claims ChatGPT and Claude cited her website more than 13,000 times in a month, and credits one thing: her entity map. She shows it on screen (a long, ugly block of brackets), explains that it tells the AI what her business does, who it is for and where the proof is, and then mentions she built a tool for her paying members that generates the block from a form so they can paste it into Squarespace or WordPress. The question people have been asking afterwards is simple. What is an entity map, and if a form can generate one, is that all I need?
The short answer is that she is right about the idea and her name for it is a good one. An entity map is a plain language label for structured data about a person or a business, written so that Google and AI can read it without guessing. The longer answer is that a block generated from a form is a start, and the part of the work that gets a person described correctly comes after the paste.
What an entity map is
Underneath every page Google and AI understand is a hidden layer (schema) that states the facts in a form a program can read: this page is about a Person called X, who runs Organization Y, wrote Book Z, has appeared on these shows, and is the same person as these profiles elsewhere. The creator's ugly block is that layer. The ugliness is the point; it is not written for people.
Call it an entity map, a schema graph or a record, it is the same thing: a set of connected nodes rather than a list of facts. The person is one node. The company is another. Each book, each talk, each show appearance is a node of its own, and every one of them points back at the person by a single permanent identifier. Structured data is the hidden graph of the web walks through why that connected shape matters more than any individual fact in it.
Her three part description is worth keeping, because it is how a buyer should think about the block rather than how a developer would:
| What she says it tells the AI | What that is in the hidden layer |
|---|---|
| What the business does | The description, knowsAbout and the services or products as their own nodes |
| Who it is for | Audience and area served, plus the topics the description commits to |
| The proof | Reviews, testimonials, press, recommendations and appearances, each a node pointing at the main entity |
That is a sound map. The reason a well built one gets cited is not magic. When an AI assistant assembles an answer about a person or a business, it prefers sources it can parse with confidence, and a page that says the same thing in prose and in the hidden layer is one it can quote without hedging. How AI search decides who you are covers the mechanics.
About the 13,000
Treat the number as the creator's own claim, which it is. There is no public counter for assistant citations, so figures like this come from a site's own server logs or from one of the AI visibility tools that estimate referrals. They can be real; a site with strong structured data and a lot of specific content is exactly the kind that gets cited heavily. But a citation count is not the same as being described correctly. An assistant can cite a page thousands of times for one fact in it and still confuse the author with a namesake, or still have nothing to say when someone types her name. Those are different problems, and the second one is the one most people actually have. What does ChatGPT say about you? is the five minute check that tells you which problem you have.
What a generator gets right
Give the creator her due. Most people with a personal site have no structured data at all, or whatever their theme or SEO plugin bolted on, which usually means a Person node with no identifier, a company node with the wrong name and nothing connecting them. A form that asks for the facts and produces a valid block fixes that in ten minutes, and ten minutes of correct structured data beats a year of none. If a generator is what gets someone to put the hidden layer on their site, that is a good outcome.
The two things a generator does well are speed and validity. It will not forget a comma, and it will not leave the @type blank. Those are real errors and they are the ones a validator catches.
Where the paste stops short
The errors that matter for a person are the ones a validator cannot see, and a generator cannot see them either, because they are errors of fact and of agreement rather than syntax. Pasting a block into the head of a WordPress or Squarespace page leaves five things undone.
One identifier, used everywhere. The block has to give the person a permanent address, something like https://yourname.com/#person, and every other node on every page has to reference that address rather than repeat the name. A generator usually produces one page's block in isolation, so the media kit, the about page and the company site each describe the person again with no shared identifier. Three descriptions, nothing connected. Person schema: what to put on your own site explains the identifier and why it is the single most important line.
sameAs that is actually you. The sameAs list is how the AI knows the LinkedIn profile, the YouTube channel and the Crunchbase entry are one person. A form will take whatever addresses you type. It will not notice that the X handle belongs to someone else with your name, that the LinkedIn address is the old one, or that the Instagram you listed is your dog's.
Prose that matches. Google and AI corroborate the hidden layer against the visible page. A fact in the block that the page never states in a sentence is a fact with no support, and a block that disagrees with the page is worse than no block. Generators write schema; they do not rewrite the home page to match, and most people never do it afterwards.
Agreement across every profile. The site is one source. The AI also reads LinkedIn, the company bio, the conference page from 2019 and the podcast show notes, and when they disagree on a name form, a job title or a date, it picks one or hedges. No block on your own site fixes a stale bio on someone else's.
The duplicate the platform already added. WordPress SEO plugins and several Squarespace and Wix templates emit their own Person, WebSite or Organization nodes. Paste a second block on top and the page now describes you twice with two identifiers. The creator's advice, paste it wherever you are, is correct as far as it goes; the step after it is to find what was already there and make yours the only one. JSON-LD explained with one example you can copy has the builder by builder notes.
A worked example
Take a consultant with a WordPress site, a LinkedIn profile, six podcast appearances and a self published book. She fills in a generator, pastes the block into the head of her home page, and validates it. Green ticks. Here is what the AI sees a week later.
- Her theme's SEO plugin already emits a Person node with her name and no identifier. Now there are two Persons on the page.
- The generated block lists her book by title, but the home page never mentions it, so there is nothing to corroborate.
- Her LinkedIn headline still says the agency role she left in 2023. The block says consultant. Two sources, two jobs.
- The six appearances are not in the block at all, because the form had no field for them, and they are not on the site either. The strongest third party evidence she has is invisible.
- Her sameAs includes a Medium address she set up in 2016 and never used. Valid, and useless.
None of this shows up in a validator. All of it shows up when someone asks ChatGPT who she is.
What the finished version looks like
- Decide the one identifier. A permanent address on your own domain. Everything else references it.
- Write the page first. Every fact you want the AI to carry is stated in a sentence on the home page: role, company, book, where you are based, how the name is pronounced. The hidden layer mirrors the page; it never leads it.
- Build the graph from the page. One Person node. Organization nodes for the companies you own on their own domains, or on yours for the ones you do not. A node per book, per show appearance, per piece of press, each pointing at the identifier.
- Remove the duplicate. Find the plugin or template schema and turn it off for the page, or make yours the only block. Confirm by searching the live source for ld+json and counting.
- Make the profiles agree. The same name form, the same headshot, the same current title on LinkedIn, X, the company bio and every directory. Email the outlets carrying the old bio.
- Add the proof as it arrives. Each new appearance or article is a new node, not a new paragraph. This is how the record grows and how citations compound.
- Check what the AI says. On day one and again at ninety days, ask Google, ChatGPT, Claude, Gemini and Perplexity who you are, and keep the screenshots.
Steps 1, 3 and 4 are what a generator approximates. Steps 2, 5, 6 and 7 are the work, and they are where the citations and the correct description come from.
Who should do which
If you are technically comfortable, have one site and one company, and are willing to do the profile cleanup yourself, a generator plus an afternoon with the guides on this site will get you most of the way. Do it. The worst outcome is a site with nothing in the hidden layer, and a generated block is far better than that.
If you have appearances, press, a book, a company or two and a name that other people share, the block is the smallest part of the job. The job is the record: one identifier, a page that states the facts, every profile in agreement, every appearance captured, Wikidata where you qualify, and someone checking what Google and AI say about you each month. That is what MeMyself builds, on your own domain, with every fact sourced and confirmed by you before it goes live. The record is the foundation; a knowledge panel, and the equivalent in ChatGPT and Claude, is what follows from it, never what you buy.