How to Position Your Business in ChatGPT and Perplexity

A prospect opens ChatGPT and asks which agency handles paid media and AI automation in their industry. One answer comes back. No page two. No ten blue links to scroll and compare. Just a name, and it is either yours or it is not.

Sit with that for a second, because it is the part most business owners still have not processed. You spent years optimizing for a format built on choice. AI answers are not. The engine writes a response and stops. If you are not inside that response, the prospect never compares you, never picks you. You were not rejected. You were never in the room.

I do not treat this as a trend I bolted onto my services because it started showing up in conference decks. Answer Engine Optimization sits inside CRAFT™, the method I work with every day: Clarity, Research, Action, Flow, Testing. Clarity defines the real business problem. Research audits the market, the competitors, the account and the data. Action ships the pages, the schema, the automations. Flow keeps the system running week to week without dead time. Testing iterates with data so results compound. AEO is not a sixth letter. It lives across all five, because being citable is a systems problem, not a content trick.

This is the playbook I actually run. Definitions first, then the audit, then the content standard, then the test protocol, then the numbers I report.

What AEO, GEO, and LLMO Actually Mean

Here is the definition I use with clients, and I keep it short on purpose because a definition that needs a paragraph is not a definition.

AI Search Optimization is the set of practices that maximize the probability that your content, your text, tables, data, definitions, insights and brand, is understandable to large language models, cited in generative engines and AI Overviews, and adapted for zero-click conversational answers. It also goes by Generative Engine Optimization (GEO), Answer Engine Optimization (AEO), and Large Language Model Optimization (LLMO).

Those three acronyms describe the same work from three angles. People get stuck arguing about which label is correct. It does not matter. What matters is the shift underneath them, and I will say my version of it plainly: SEO in 2026 is no longer about rankings or keyword tricks. It is the discipline of orchestrating digital systems and assets to maximize visibility, demand and trust in every space where a person searches, asks or discovers, including AI engines.

One more thing I tell every client before we start: only a very small percentage of content on the internet is genuinely cited, linked and used as the foundation for LLM answers. The sources that get picked are the most understandable, the deepest, the most trustworthy and the most differentiated. That is a high bar, and it is also good news, because it means the bar is reachable by a small business with real expertise and unreachable by a content farm with a budget.

Why ChatGPT and Perplexity Are Not the Same Channel

Most people treat "AI search" as one thing. I do not, and neither should you.

How ChatGPT retrieves and cites

In standard chat, ChatGPT answers from training data alone. It pulls live web results only when its search tool fires, either automatically or because the query obviously needs current information. When it does search, it reads a handful of selected pages and writes an answer with citations drawn from that content. You do not control which pages get selected, and the selection criteria are not the signals Google uses to rank.

How Perplexity approaches answers

Perplexity is search-first. It queries the web before generating almost every answer and shows inline citations tied to specific claims. That makes source selection more transparent, and it also makes it more winnable. Any page that clearly answers the query can get cited regardless of domain authority. Perplexity leans toward pages with a directly extractable answer, which is frequently not the biggest name in the category.

Why you have to test both separately

This is the part I want you to underline. Analyses published in 2026 covering hundreds of millions of citations report that only around 11% of the cited domains overlap between ChatGPT and Perplexity, and that brand mention rates differ sharply from one engine to the other.

Eleven percent. For practical purposes that makes them two different channels with two different source pools. Winning one tells you almost nothing about the other. Every time someone shows me a screenshot of ChatGPT recommending them and concludes they are "visible in AI," my first question is what Perplexity says. Nine times out of ten, nobody checked.

Two priorities fall out of this. First, entity clarity: the model has to recognize who you are, what you do, and that you are a distinct, verifiable business and not a name collision with a competitor. Second, answer extractability: your content has to be formatted so a model can lift a clean, accurate answer straight off the page. Chase keyword rankings alone and you miss both.

Visible Brand vs Invisible Brand

Before tactics, I run this diagnostic. It takes ten minutes and it usually ends the debate about where the problem is.

Read the right column honestly. If three or more lines describe your site, no amount of schema markup will save you, because the problem is not technical. You are producing content that has no reason to be cited.

The beginner mistakes I see repeat in almost every audit are the same four: picking topics by search volume alone, confusing "published" with "distributed," ignoring intent and thinking only about Google, and using AI to copy rather than to differentiate and go deeper. Publishing is not distributing. A business that publishes and waits wastes most of the value of every piece it makes.

Writing Content That AI Engines Actually Quote

Now the operational part. This is where I spend most of my time with clients, because schema is a weekend of work and content is the ongoing discipline.

The content types AI cites most

From my own SEO playbook, the formats that get picked up, in order of how reliably I see them work:

• Clear, precise definitions. "What is X" plus examples.
• Comparison tables. Heavily used for "vs," "best for," specs, pros and cons.
• Downloadable resources with data, glossaries, proprietary frameworks, calculators.
• Structured FAQs with short answers, each linked to a deeper dive.
• Cases, storytelling, verifiable practical examples. The "we did this" content.
• Product opinions and reviews with evidence, photos where you can.

And the most actionable line in my whole playbook: AI scrapes tables before it scrapes traditional text. If you take one thing from this article, take that. Add your own tables. Comparisons, pros and cons, steps, metrics, alternatives, technical specs. I have watched pages start getting cited after nothing changed except one well-built comparison table.

Bad versus good, side by side

Easier to show than to explain. Same topic, two versions.

Bad:

"Factoring is a financial tool. It helps companies improve their liquidity."

Good:

"Factoring is a financial solution that lets a company collect the value of its outstanding invoices in advance, optimizing cash flow. Example: if your business has invoiced $100,000 and needs immediate liquidity, factoring lets you assign that invoice to a provider who pays you a percentage now and holds the rest until the invoice is collected."

Then you add a comparison table, a FAQ block, and a real case underneath.

The bad version is true and useless. Nothing inside it can be extracted, because there is nothing a model can hand to a person as an answer. The good version defines the mechanism, names the actors, and includes a concrete number. A model can lift that second paragraph whole and it stands on its own with no surrounding context. That is the test.

The formatting rules I enforce

• Definitions: open a guide or a key topic with a synthetic one to two sentence definition, then examples.
• Structure: put "the essentials" in a block before the long explanation. Give the ideal zero-click answer up front, and put the deep dive below it to earn the click from the advanced reader.
• FAQs: answer high-volume questions, pain points, funnel objections and queries you have actually seen in LLMs, with no throat-clearing. Cover the minimum set for any topic: what it is, how it works, advantages, disadvantages, alternatives, and who it is not for.

That last one is my favorite, and almost nobody does it. "Who this is not for" is a section most businesses are terrified to write. Write it. It is the single clearest signal of honesty you can send to a reader and to a model, and it filters out the leads that were going to waste your sales team's time anyway.

Entity Clarity: Structured Data as a Fingerprint

Schema will not force a model to cite you. What it does is strengthen the three conditions that raise your odds: entity clarity, verification, and extractability.

Organization markup on your homepage and about page defines you as a distinct entity: name, URL, logo, contact. LocalBusiness adds address, phone, hours, service area. FAQPage gives engines question and answer pairs they can lift directly into a response. HowTo does the same job for procedural content with numbered steps.

The property most people skip is sameAs. It links your entity to authoritative external profiles: your Google Business Profile, your LinkedIn, relevant industry directories. When a model finds multiple references across the web that all point back to the same entity, confidence in citing you goes up. Inconsistent naming across platforms works directly against you, because the model cannot confirm whether you are one business or three loosely related listings.

Fix your schema this month and you close the gap between "technically online" and "recognizable as an entity."

Off-Site Authority: Where the Citations Actually Come From

On-site work defines your entity. Off-site work gets it trusted, and trust is what earns the citation.

Let me be careful about attribution here, because the sharpest public data on this is not mine. Neil Patel reports that over the last 18 months rankings have stayed relatively stable while organic clicks have fallen hard, especially on queries most affected by AI Overviews. According to data published by NP Digital, industry figures show more than half of Google searches now end without a single click to a website, and in their own paid search testing AI Overviews cut click-through rates by more than 50% on affected queries. Patel's line for it is that the SERP used to be a doorway and is now a destination.

The finding of his I keep coming back to: Neil Patel has written that when his team analyzed 48 sites in their portfolio, organic traffic was falling while direct revenue was rising. People were seeing the brand inside the AI answer and then going straight to the site, skipping the search click entirely. If you are only watching sessions, that looks like a loss. It is not.

NP Digital also published a client case I find more instructive than any statistic: a brand that did not rank in the top ten for its best keywords turned out to be the most cited brand in its category across AI systems, because it had spent three years investing in PR, podcasts, community and creator partnerships. They were not optimizing for AI. They were building brand presence, and the models rewarded them for it.

Here is my own read on top of that, and this is where I build rather than borrow. Large language models look for consensus among sources. The more places describe a business as a category leader, the more the model treats it as the default answer. That has two practical consequences.

First, third-party presence beats owned presence for branded and category queries. Reviews, listicles, forum threads, industry publications, YouTube. If your business appears only on your own domain, the model has exactly one source, and one source is not consensus. If it appears across a dozen credible third-party references, the model is no longer taking your word for it.

Second, reviews function as liveness signals. I am not going to give you a number here, because the numbers floating around on this are not verifiable and I do not use figures I cannot stand behind. Qualitatively, what I see is consistent: businesses with complete, actively maintained profiles across two or three relevant review platforms show up more often in AI answers than businesses with one stale listing. A profile with recent activity and real responses reads as a live business. A profile that has not moved in two years reads as a question mark. Pick the two or three platforms that matter in your industry, keep them complete and consistent, and respond to reviews. The good ones and the hard ones.

Analyze the SERP, Not the Keyword

This is a research habit and it is where most AEO programs go wrong before they write a word.

Do not read the keyword. Read the result page. Look at titles, snippets, People Also Ask, video presence, featured snippets, and the pages actually ranking. If YouTube, marketplaces, forums and AI results are showing up high, the intent is not purely informational and a blog post is the wrong asset.

Then do the thing almost nobody does: search conversationally. Ask the question the way a real person would ask it, directly inside ChatGPT, Perplexity, Gemini and Claude. Look at the first twenty real results across Google, YouTube, LinkedIn and TikTok for each cluster. Detect the dominant format. Note who is getting cited by AI and by snippets. That list of cited sources is your actual competitive set for AEO, and it is usually not the same list as your Google competitive set.

On keyword selection I have one rule I do not bend: never prioritize by volume alone. Prioritize queries tied to commercial intent, to decisions, or to the doubts that block conversion. Map every keyword to a funnel stage and a specific page type. If you cannot turn it into an action, it is not worth the slot.

Topical Authority: Go Deep on Two to Four Things

Topical authority is the depth and coherence a site demonstrates about a specific topic and its subtopics, in the eyes of users, search engines and AI models. The goal is to become the reference source on a pillar, at a depth no competitor can easily match.

The number I work with: two to four core topics. New site, make it one or two. I have never seen a site win AI citations by covering fifteen topics adequately, and I have repeatedly seen small sites win by owning two topics completely.

Two rules go with it. Measure by cluster, not only by page, because a cluster can be winning while three of its pages look flat. And only create a new page if there is a clear intent and a clear value an existing cluster does not already cover. Audit your top similar pages every quarter, then merge, redirect or delete whatever is cannibalizing.

E-E-A-T and the Anti-Generic Test

E-E-A-T is not a checkbox. It is four separate things and most businesses have one of them.

• Experience: demonstrate you have actually solved the problem or used the product. Own cases, tests, honest reviews, stories, photos, video.
• Expertise: a demonstrably expert author. Credentials, studies, certifications, prior publications, a verifiable background.
• Authoritativeness: your reputation off your own site. Links, mentions, citations, external reviews, press presence, linkable assets.
• Trustworthiness: security, transparency, contact information, clear policies, evidence, sources, exact data, and the legal architecture of About, Contact, Privacy and Terms.

For local and service businesses, what moves the needle is your own cases, testimonials with a name and a city, a verified Google Business Profile, real photos, and your actual team on the page.

Now the test I apply to every piece before it ships, word for word from my own playbook: never publish without adding a real example, a story, unique data or an internal point of view. Audit the text with human judgment and ask whether any generic writer could have written the same thing. If yes, redo it and add experience. Without real experience, you are invisible to both Google and the models.

One more trust lever that costs nothing: make your process and your limits transparent. Publish what you do not do and what you cannot guarantee. Show real criticism and respond to it openly. Every competitor's site claims the same things. The business that publishes its own boundaries is the one that sounds like it has actually done the work.

The 10-Point AI Search Optimization Checklist

This is the checklist I run against a page before it goes live. Ten questions, all answerable yes or no.

1. Does every key term have a clear, demonstrable definition?
2. Does the content include tables, lists, or structured and updated data?
3. Do the FAQs cover more than 85% of People Also Ask plus real AI queries?
4. Are there real cases, real experience, or an original point of view?
5. Does every primary and secondary entity appear at least once?
6. Are sources cited and linkable?
7. Is the content unique and citable for AI engines and aggregators?
8. Is the authorship, the date, and the context of the experience easy to find?
9. Is the format optimized for snippet, conversation, and AI answer?
10. Could this piece be cited, without hesitation, as a primary reference?

If question ten is a no, the other nine do not matter. Go back and fix the content, not the markup.

How to Test Whether ChatGPT and Perplexity Recommend You

Everything above stays invisible until you validate it. This is the protocol, and it is deliberately low-tech, because a spreadsheet you actually maintain beats a dashboard you open twice.

Build the prompt set. Ten to twenty prompts mirroring real buyer queries across four types: category questions ("best [service] for [use case]"), problem-solution questions ("how do I solve [problem]"), comparison queries ("alternatives to [competitor]"), and direct branded queries ("what is [business name]").

Run each prompt three times in a clean or incognito session, because model outputs vary run to run and a single run tells you nothing.

Test ChatGPT and Perplexity separately. Given the roughly 11% domain overlap between them, blending the results into one number destroys the signal.

Log six fields per run: whether your brand appears, its position in the answer, whether a source URL from your domain is cited, which competitors are mentioned, sentiment toward your brand, and whether the description of your business is accurate.

That last field catches problems the other five miss. I have seen a business get cited consistently with a description of services it stopped offering two years ago. Being cited wrong is its own category of problem, and you only find it if you are reading the answers instead of counting them.

Separate branded prompts from unbranded prompts in your log. Branded prompts tell you whether the model knows your name. Unbranded prompts tell you whether you have category-level recognition, which is the one that generates new business.

The Five Numbers I Report

Testing tells you where you stand. These tell you whether you are improving.

Note that mentions and citations are separate lines, and that mentions come first. A model naming your business without linking it still sends buyers to you, they just arrive by typing your name into the address bar. That is exactly the pattern in the NP Digital portfolio finding I cited above, and it is why I no longer accept a traffic drop as sufficient evidence that a program is failing.

This measurement layer is the Testing stage of CRAFT™ and I do not run an AEO engagement without it. If you cannot count your citations, you cannot improve them, and you certainly cannot prove to a CEO that the work is moving.

Frequently Asked Questions

What is Answer Engine Optimization? AEO is the practice of structuring your content, your entity data and your off-site presence so that AI engines can understand, extract and cite you as the answer to a question. Unlike classic SEO, it does not target a position in a list. It targets being included in a single generated response.

How long does it take to show up in AI answers? Entity and schema fixes can register within weeks. Citation authority, which depends on third-party mentions and accumulated content depth, moves on a scale of months. The NP Digital case I referenced above involved three years of PR, podcasts and community work before the brand became the most cited in its category, which is a useful reality check on anyone promising results in thirty days.

What are the advantages over traditional SEO? Domain authority matters less than it does in ranked search, which means a focused small business can outcompete a large one on a specific question. Extractability and genuine expertise are the levers, and both are within reach of a team of any size.

What are the disadvantages? Attribution is harder. A large share of the value arrives as unlinked brand mentions and direct traffic, so your analytics will understate the result. Tooling is immature and changing fast. And you cannot control which pages an engine selects, only how likely yours is to qualify.

Do I still need traditional SEO? Yes. Technical health, indexation and quality content remain the entry requirements. Traditional SEO is now the floor rather than the ceiling. Skip it and you will not be in the retrieval pool for any engine to consider.

Who is this not for? If you have no genuine expertise, no real cases and no intention of publishing an original point of view, AEO will not work for you and I would rather tell you that now. The whole mechanism rewards content nobody else could have written. If your plan is to generate volume with AI and hope the models pick you up, you are producing exactly the material every engine has learned to skip.

Start Here, This Week

I am not going to end this with a paragraph about how the future is arriving. You already know that, or you would not have read this far.

Instead, here is the sequence I would run if this were my own business, in order, starting Monday.

Day one: write ten prompts your buyers would actually type, run them three times each in ChatGPT and in Perplexity, and log what comes back. Do not fix anything yet. Just look.

Day two: pick the single question your business answers better than anyone in your category. One question. Write the definitive answer to it, opening with a two-sentence definition, including one comparison table you built yourself, and closing with an FAQ that includes who it is not for.

Day three: fix your Organization and sameAs schema, and make your name identical everywhere it appears online.

Day four: list every third-party site where your business could credibly be mentioned and start one conversation.

Day five: put the six log fields in a spreadsheet and put a recurring calendar invite on it for the same day next month.

That is a week. Most of your competitors will spend the same week arguing about whether AI search is worth optimizing for, and the models will keep building consensus without them in it.

If you want me to run the audit instead, that is what I do. I will tell you which engine already knows you, which one does not, whether the description it gives of your business is even correct, and what it will take to fix it. You will get me, not a junior, and you will get the real answer rather than the flattering one.