Skip to content

Article 7 min read

The SEO Master Prompt: Four Steps and the Entity Layer Almost No Prompt Has

by Daniel Sócrates published

The SEO Master Prompt is a four-step prompt that makes the AI map a topic’s search intent and entities before it writes a single line. What sets it apart from an ordinary prompt is the entity layer. Instead of asking for a text about a keyword, it makes the machine declare which thing it is talking about. It also has to state that thing’s attributes and values, and how it relates to the other things in the subject.

This prompt began in a video on my Instagram, where people commented to get it. Copy it, paste it into the AI you use, fill in three lines at the end, and run it. Rather than a loose list of requests, it is four steps in a fixed order. The AI has to complete the first two before it writes any of the text.

The layer almost no prompt has

Almost every SEO prompt tells the AI to write about a topic. The text comes out polished, covers the keyword, and goes nowhere.

What is missing is the entity layer.

Entities are how Google stores the world. It works with things: people, companies, products, tools, places, concepts. Each of those things has attributes and values. What it is, what it does, who it serves, where it operates, what it costs. This is called EAV, Entity-Attribute-Value. 1

When your text names the right entities, explains how they relate, and keeps the values consistent, Google understands what you are talking about. That understanding is what gets the content retrieved, and then cited by ChatGPT, Perplexity, Gemini, and AI Overviews.

Here is my reading of what this means in practice, and it is a thesis rather than a fact from official documentation. Google does not rank text. Google ranks the answer to the intent. 2 The entity layer is what tells the machine which thing you are answering about.

Steps 1 and 2 of the prompt exist for exactly that. The AI maps intent and entities first, and only then does it write.

Before you run it: the check that can void everything

One warning before you use it. This prompt works on the text, and text only gets cited if it gets found. If your page is not indexed, no step below works. No amount of good writing makes up for a page Google does not have.

Check this first in Google Search Console, in the Pages report under Indexing. It is the same report I use in a technical audit, and it tells you within a minute whether the problem is content or access.

The Master Prompt

Copy the whole block, from start to finish.

bloco para copiar
You are an expert in semantic SEO, Entity SEO, and GEO (Generative Engine Optimization).
Run the four steps below in order. Show me the results of steps 1 and 2 before you write
the text.

STEP 1: INTENT
1.1 State the real search intent behind the topic: what the person wants to solve when
    they type it.
1.2 Classify the intent level. Base: they have the problem and do not yet know the name
    of the solution. Middle: they already know the name and are comparing options. Tip:
    they are searching for the product and are ready to buy or hire.
1.3 List the 3 questions this person asks right after the main question.
1.4 Choose one level only and write for it from the first paragraph to the last.

STEP 2: ENTITY MAP
2.1 Name the central entity of the topic and its type: person, company, product, tool,
    place, concept, event, or creative work.
2.2 If the name is ambiguous or has a well-known namesake, disambiguate it right in the
    first paragraph: type, field of activity, and context.
2.3 List the attributes and values of the central entity. Cover what it is, what it does,
    who it serves, and where it applies. Add what it costs when that makes sense, and what
    sets it apart from the alternatives. Flag with [CONFIRM] every value you cannot back up.
2.4 List 5 to 10 related entities the subject needs in order to be complete: tools,
    competitors, places, standards, key people, and adjacent concepts.
2.5 For each related entity, write one sentence stating how it relates to the central
    entity.
2.6 Use the full official name of each entity on first mention.

STEP 3: WRITING
3.1 Open with a summary paragraph of 2 to 3 lines that fully answers the intent from
    step 1. Write that paragraph to be cited on its own, outside the text.
3.2 Organize the content under subheadings that are the questions from step 1.3, in the
    order the person would ask them.
3.3 Write paragraphs of 3 to 4 lines, each one answering a whole idea. Someone who reads
    only that paragraph has to understand it.
3.4 Use the entities from the step 2 map throughout the text, and make the relationships
    between them explicit. No entity goes in unconnected.
3.5 Cover the semantic field of the topic: variations, co-occurrences, and the terms that
    always show up together when someone really knows the subject.
3.6 Use numbered lists for steps, criteria, and comparisons. Each item has to make sense
    on its own.
3.7 Put the sentence that sums up each block in bold.
3.8 Attribute every strong claim to whoever makes it: a person with their job title, a
    company, or a study with author and date.
3.9 No vague sentences. Use data, numbers, tool names, and concrete examples.
3.10 Close with a Key points section of 3 to 5 numbered items. Write each one as a
     complete answer that makes sense when read outside the text.

STEP 4: REVIEW
Review the text and fix it before you hand it to me:
4.1 Does the text answer the intent of the chosen level from start to finish?
4.2 Does the central entity appear under its official name, and is it disambiguated?
4.3 Is every attribute value consistent with what is already published about this
    entity? Point out any divergences you find.
4.4 Does the summary paragraph work as a self-contained answer, outside the text?
4.5 Is any strong claim left without an owner?
4.6 List what you fixed.

INPUT
Topic: [write the topic of the content here]
Central entity: [your brand, your product, or the concept of the topic]
City or region: [fill in only if the business is local]

Why each step exists

A prompt with no reasoning behind it is superstition. Each item in the list above solves a specific problem, and you should know which one.

ItemWhat it solves
1. Intent and levelIntent has levels, and it is the same person searching differently as they learn. The right topic at the wrong level brings visits and no customers. Choosing one level is what gives the whole text its focus.
2. Entity mapThis is the step almost nobody takes. The AI needs to know which thing it is talking about before it talks. Without the map, it writes about the word instead of the entity.
2.3 Attributes and valuesGoogle’s Knowledge Graph stores your entity as triples: entity, attribute, value. Whatever system answers a factual question reads those triples. Declaring the attribute in one text is not enough, because the value has to match everywhere you appear.
2.4 Related entitiesGoogle understands a subject through the things that show up alongside it. A text that names the right neighboring entities covers the whole subject. A text that stays on the keyword covers one slice.
3.1 Summary paragraphThis is the block that AI Overviews and Perplexity pull as the answer. Written to work on its own, it travels outside your site carrying your name.
3.4 Explicit relationshipAn entity dropped in without context teaches the machine nothing. The relationship between two entities is what becomes new information in the Knowledge Graph.
3.8 AttributionThe AI needs to know who is making a claim before it can cite anyone. Data with no owner becomes loose information, and the citation goes to someone else.
4.3 ConsistencyThis is the part most people miss. If your site says one thing and your profile says another, the machine is left unsure which value is right. An entity in doubt does not become an answer.

How to use it

Fill in the three lines of the INPUT block and run it. The AI will first return the intent diagnosis and the entity map. Read that map before you accept the text.

If any value comes back flagged with [CONFIRM], check it yourself before you publish. A wrong value, once published, is worse than no value at all, because it starts to diverge from everything you already have online.

Three examples of how to fill it in:

  1. Topic: how much it costs to put solar panels on a home in 2026. Central entity: residential solar power. City: Phoenix.
  2. Topic: a comparison of ActiveCampaign and HubSpot for small businesses. Central entity: ActiveCampaign.
  3. Topic: what to check before you commit to a dental implant. Central entity: your clinic.

How to measure

Use the prompt on one piece of content per week for 90 days. Then compare the results in Google Search Console.

Compare on the Pages tab of the Performance report instead of the Queries tab. The Queries tab sees about a third of real clicks, because Google anonymizes a large share of searches. 3 And compare clicks rather than impressions or average position.

The case in that video took months of content published this way. Nobody can promise you the same number, and anyone who promises it is selling something.

If you want to understand entities in depth

This prompt applies the entity layer inside a single text. It works, and it is the surface.

The bigger work happens off the page. You build your entity’s official home and keep its attributes consistent everywhere it appears. And you get sources you do not control to confirm who you are. That part has an order, and skipping a step costs months.

I wrote an entire book about this. It is published in Portuguese as SEO de Entidade, and its English edition, AI Search Optimization, is scheduled for September 2026. The subtitle says what the book delivers: how Google and AI decide who to cite, and what to do about it. It has 319 pages on the mechanisms, the patents, and the step-by-step method. It covers the four gates of citation, the Inverted Intent Pyramid, and the map for building your entity’s official home. 4

Here is one fact about the book itself. It came out on August 12, 2026. Four days later, Google showed an AI Overview for the search “SEO de entidade” (Portuguese for entity SEO). The first source it cited was the book’s page. No ads and no paid links, just the method applied to itself. Search results change over time, and this is the snapshot from August 16, 2026. 5

The book is mine, and it is not free. It lives at seodeentidade.com, which gathers every place to buy it. That page is in Portuguese, like the edition you can buy today.

If you want to learn SEO beyond the basics, follow me at @danielsocrates.seo. I teach practical SEO there every day, in Portuguese.

Frequently asked questions

Do I have to pay anything to use the prompt?

No. The whole prompt is on this page, with no signup and no email required. The book is a separate product, mine and not free, and it covers the work that happens outside the text.

Which AI does it work in?

Any AI that accepts long instructions: ChatGPT, Claude, Gemini, Perplexity. The prompt does not rely on a proprietary feature of any of them. What it does rely on is your page being indexed, and that comes before any model.

Why does the AI have to show steps 1 and 2 before it writes?

Because that is where it goes wrong without you noticing. If the intent level is wrong, the whole text misses its target and still looks good. If the entity map comes back shallow, the AI writes about the word instead of the thing. Seeing both steps first gives you a chance to fix it cheaply.

How long until I see results?

I do not have that number to give you, and be wary of anyone who does. The protocol is one piece of content per week for 90 days, comparing clicks on the Pages tab. What I can tell you is what to measure and where to look.

Notes and sources

  1. The Entity-Attribute-Value model, the way the Knowledge Graph stores each thing: the entity, the attribute, and the value of that attribute. Developed in the book SEO de Entidade, published in Portuguese.
  2. The author’s reading, and not a statement from Google’s official documentation.
  3. The query dimension of Google Search Console covers around 31% of real clicks, because queries are anonymized. The page dimension covers 100%. That is why the performance comparison is made on the Pages tab.
  4. SEO de Entidade, Daniel Sócrates, Clube de Autores (a Brazilian self-publishing platform), August 12, 2026, 319 pages, 16 x 23 cm format, ISBN 978-65-266-8059-9.
  5. Record of the AI Overview citing the book’s page as the first source on a search for “SEO de entidade”, captured on August 16, 2026, four days after launch. A results panel is a snapshot of one moment, and it changes over time.

Want to see SEO Genome more often on Google? Add it as a preferred source.

Daniel Sócrates

Founder and CEO of SEO Genome

Author of AI Search Optimization. Writes about what the company applies in client work.

danielsocrates.com.br

Let us look at what the machine understands about your company

A diagnostic conversation, with no slide deck. You bring your website address. We bring what Google and the AI assistants already know about it today.

Talk to us on WhatsApp