(214) 272-7034 
info@fasthippomedia.com
ChatGPT decides whether to show you before it decides what to say about you, and that first decision lives in your robots.txt file. After that, GEO is five things: check you are visible to the crawlers, audit what AI currently says about you, publish facts a machine can verify, earn third-party mentions models can trust, and structure everything to be summarized. The 30-day sprint at the end is the short version. This section is also a demonstration: it is written to be lifted whole, which is the skill the rest of the post teaches.
Search outgrew the blue links
Buyers are not only Googling. They ask assistants to explain, compare and recommend, and tools like ChatGPT and Google’s AI Overviews compose an answer in seconds, citing a handful of sources they trust.
If your brand is not in those answers, you are absent at the exact moment the decision gets made. Not ranked low. Absent, with no page two to be on.
This guide covers Generative Engine Optimization, or GEO: the practice of shaping how AI systems understand, summarize and cite your brand. It sits across PR, SEO and data strategy rather than inside any one of them, which is why it tends to fall between teams. By the end you will have a playbook and a 30-day plan you can start on Monday.
GEO is one of three disciplines that overlap more than they compete, and we set out how separately. This page is the deep version of the third one. Running it for clients is what we do, and everything below is the working method rather than theory.
Step zero: check you are not switched off
Before any of the strategy, one thing decides whether the rest can work, and almost nobody checks it.
OpenAI runs four separate crawlers, and they do different jobs. Per :
OAI-SearchBotsurfaces sites in ChatGPT’s search resultsGPTBotcrawls content for model trainingChatGPT-Userhandles fetches a user triggers inside ChatGPTOAI-AdsBotvalidates ad landing pages
The sentence that matters, verbatim: “OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features. Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.”
Read that again with your own robots.txt in mind. Through 2024 and 2025, a great many sites blocked AI crawlers on principle to keep their writing out of training data. That is a completely defensible position. The trouble is that the block was often written broadly, and blocking the training crawler and blocking the search crawler are two different decisions with two different consequences. One keeps your content out of a model. The other removes you from ChatGPT’s answers entirely.
Perplexity is structured the same way. describes PerplexityBot as “designed to surface and link websites in search results on Perplexity”, and states plainly that it is “not used to crawl content for AI foundation models”. A separate agent, Perplexity-User, handles fetches triggered by a user’s question.
So the first move in any GEO programme is thirty seconds long. Open and read it.
Here is ours, in full, as of today:
User-Agent: *
Allow: /
Disallow: /wp-admin/
Disallow: /author/
A wildcard allow, an admin directory blocked, an author archive blocked. Every search crawler named above is permitted. That is what a clean one looks like.
If yours names GPTBot or PerplexityBot in a Disallow, you have a decision to make rather than a bug to fix, and it is worth making deliberately: training and search are separate, and you can allow one and refuse the other. If it disallows OAI-SearchBot, you are not doing GEO. You are opted out of it.
What GEO is, and what it is not
SEO earns a position: Someone searches, results are listed, you occupy one of them. The unit is a page, and that is what SEO works on.
GEO earns a mention: Someone asks, an answer gets composed, and either your expertise is part of it or it is not. The unit is your brand as an entity: who you are, what you do, what you are known for, and how consistently that is stated across everything a model can read.
That difference drives everything else. Page-level optimization moves a page. Entity-level work moves what the system believes about you, and that belief is assembled from your site, your profiles, your press coverage, your listings and everywhere else your name appears with facts attached.
It is closely related to , which targets the direct-answer slot. The techniques overlap heavily. The distinction worth holding is the surface: an answer box quotes one source, while a generative answer synthesises several and names a few.
Why earned authority does more work here
In classic SEO, what you publish about yourself carries real weight. In generative answers it carries less, because a system assembling an answer has an obvious preference for a claim it can verify in more than one place.
A statement about your business that appears only on your own site is a claim. The same statement appearing in a trade publication, a directory, a conference programme and a customer’s case study is a fact. Models are built to prefer the second, and that preference is the whole reason GEO pulls PR into a conversation that used to be purely technical.
This is also why GEO cannot be delivered by a content team alone, and why the stops being a separate budget line and starts being an input to search visibility.
The GEO playbook
Seven steps after step zero. They run in this order for a reason: each one makes the next cheaper.
1. Run an AI Visibility Audit
Ask the major tools what they currently say about you, and write down the answers rather than reacting to them.
Four prompt types, run on ChatGPT, Google’s AI Mode, Perplexity and Copilot, signed out:
- Brand: “What is [your company] and what do they do?”
- Category: “Who are the best [your service] providers in [your city]?”
- Comparison: “How does [you] compare to [competitor]?”
- Adversarial: “What are the downsides of using [you]?”
What you are recording: whether you appear at all, what facts the tools state about you, which of those facts are wrong, which sources they cite, and who else gets named. That last one usually surprises people, because the competitive set a model has is often not the one you think you are in.
What comes out of it: a list of factual errors to correct at source, a list of questions where you are absent, and the list of domains the tools are actually trusting in your category. That third list is your PR target list, handed to you for free.
Run it again monthly. One run is an anecdote, three months of runs is a trend.
2. Build a machine-readable fact base
Every fact about your business should be stated somewhere a machine can parse it, not implied by your design.
The core set: legal and trading name, what you do in plain words, where you operate, hours, pricing where you publish it, credentials and licences, key people with their actual expertise, and the questions customers ask before buying.
Then state them in structured data on the pages they belong to, generated from your own fields rather than pasted in by hand. Pasted markup drifts the moment the copy changes, and drifted markup is worse than none because it makes your own page an unreliable source. The implementation side of this is work.
The test for this step is not whether the schema validates. It is whether a stranger reading only your structured data could describe your business correctly.
3. Align PR, SEO and content into one strategy
Three teams usually own three parts of the same problem and report on it three different ways.
PR earns the third-party mentions that make your claims verifiable. SEO makes the pages readable and linked. Content produces the substance both of them need. In a GEO programme they share one target list: the questions you want to be the answer to, and the domains you want saying your name.
The practical change is small and awkward: one shared list, one meeting, and PR success measured partly on whether the coverage says the things you need said, not only on where it landed.
4. Design for summarizability
Write so a passage can be lifted whole and still be true and useful.
That means the answer comes first and the context follows, one idea per paragraph, headings phrased the way someone would ask, and specifics rather than adjectives. “Most projects complete in two to three weeks” is liftable. “Fast turnaround times” is not.
The test: take any paragraph out of your page, show it to someone who has not read the rest, and see whether it stands up. If it needs the paragraph above it to make sense, a model has nothing to quote.
The TL;DR at the top of this page is the demonstration. It is written to be lifted whole.
5. Publish work a model has reason to trust
Thought leadership is a term that has been emptied out, so here is the operational version: publish things that contain information a reader cannot get anywhere else, and that another publication would have reason to cite.
Original data from your own operations. Corrections of things widely believed and wrong in your field. Methods explained in enough detail to be followed. Real numbers with sources attached.
The reason this belongs in a GEO playbook rather than a content-marketing one is mechanical. Citable material earns third-party mentions, third-party mentions make your facts verifiable, and verifiable facts are what a system reaches for when composing an answer. Step five feeds step three, which feeds step two.
6. Measure what you can actually measure
Be careful here, because this is where GEO reporting most often becomes fiction.
What is genuinely measurable: referral traffic from AI tools, visible in analytics by referrer. Your monthly audit results, recorded consistently. Impressions in Search Console’s generative AI report, where your property has it. Mentions and citations you find manually.
What is not measurable: how many people saw your brand named inside an AI answer. Nobody can measure that, including us. Any report showing it as a precise number is estimating, and should say so.
The honest scoreboard is the audit run monthly, logged in the same format each time, plus AI referral traffic as a secondary signal. It is less satisfying than a dashboard and it has the advantage of being true. Where this fits into visibility across every surface rather than just one is covered in .
7. Close the loop
The output of each month’s audit is the input to the next month’s work.
Fix the factual errors at source rather than complaining about them. Fill the gaps where you were absent by publishing the answer. Feed the trusted-domain list to whoever owns PR. Then run it again.
Most of what makes this work is not clever. It is doing the loop enough times that the corrections compound.
The roles this creates
Worth saying out loud, because the work does not fit existing job descriptions and tends to fall down the gap between them.
Someone owns the entity: One person responsible for every factual statement about the business being consistent everywhere it appears. Today this is usually nobody, which is why hours are wrong in three places.
Someone owns the audit: Runs it monthly, logs it the same way each time, and escalates what changed. This is a two-hour job that most organisations do zero times.
Someone bridges PR and search: Not a new hire in most cases, but an explicit responsibility: making sure earned coverage says the things that need saying.
None of these need to be full-time. All of them need to be somebody’s.
Where this usually goes wrong
Treating it as an SEO task: GEO needs PR and operational facts. An SEO team alone will do steps two and four well and cannot do step three.
Optimising pages while the entity is a mess: If your hours, services or location disagree across your site, your profiles and directories, no amount of on-page work fixes what a model believes.
Chasing a number nobody can measure: See step six. If someone sells you a precise AI visibility percentage, ask how it is calculated.
Publishing more instead of publishing better: Generative answers cite few sources. Being the eleventh page on a topic gains nothing.
Blocking the crawlers by accident: Step zero. It is the cheapest possible mistake to fix and the most expensive to leave.
Your first 30 days
Week 1: Read your robots.txt. Run the AI Visibility Audit across all four prompt types and all four tools. Write down every factual error and every question where you are absent.
Week 2: Fix the facts at source. Correct your profiles, your listings and your site so they agree. Add or regenerate structured data on your key pages.
Week 3: Take the three questions you were most obviously absent from and answer them properly, written to be summarized. Pitch two or three earned-media opportunities from the trusted-domain list your audit produced.
Week 4: Run the audit again. Compare it to week one. Log both in the same format, because month two is where this starts being a trend rather than a snapshot. Then decide the next three questions.
Thirty days does not make you the answer. It tells you where you stand, removes the errors working against you, and gets the loop running. This is a compounding programme, and it needs a spin-up before it compounds. Getting the same material onto every surface at once is what exists to do.
Start with the file
If you do one thing after reading this, open .
It takes thirty seconds, the answer is binary, and it is the only step in this entire playbook where the outcome can be that none of the rest matters.
Then run the four prompts and write down what the tools say about you. That is the first hour of every GEO engagement we run, and it is the part most businesses find genuinely uncomfortable, because the tools tend to be confidently wrong about something that matters.
If you would rather we ran it, and we will start there.
Frequently Asked Questions
GEO is the practice of shaping how AI systems understand, summarize and cite your brand, so that when a tool like ChatGPT composes an answer, your expertise is part of it. It optimizes your brand as an entity rather than optimizing individual pages.
SEO earns a position on a results page and works at the level of a page. GEO earns a mention inside a composed answer and works at the level of your brand, which means it depends on structured facts and third-party verification as much as on your own content.
Because buyers increasingly ask AI tools to explain, compare and recommend before they ever reach a results page. If those answers do not include you, you are absent at the point the decision is made, and there is no second page to be found on.
Yes, and the referral traffic is visible in your analytics. What is not measurable is how many people saw your brand named in an answer without clicking, so treat any precise figure for that with suspicion, including ours.
Yes. OpenAI documents a separate crawler, OAI-SearchBot, for surfacing sites in ChatGPT’s search results, and states that sites opted out of it will not be shown in ChatGPT search answers. Sites that blocked AI crawlers to stay out of model training sometimes blocked this one too. Check your robots.txt.
Read your robots.txt first. Then ask the major AI tools what they say about your brand and your category, log the errors and the gaps, and fix them at source with structured facts on your site, consistent profiles, and earned third-party mentions.
