GEO: how to get ChatGPT and Perplexity to recommend your brand

Ask ChatGPT or Perplexity "what's the best project management tool for a small agency?" or "which Canadian outdoor brand has the easiest returns?" and you'll get a short list of names, with reasons. Being on that list is the new page one. Generative Engine Optimization (GEO) is the discipline of earning a place on it, and making sure what the assistant says about you is true.

This playbook explains how AI assistants decide what to recommend, how to measure where you stand, and the specific changes that tend to move the needle. It's deliberately practical: no guarantees, no magic prompts, just the work.

Why this matters now

The behaviour shift is well documented. In McKinsey's AI Discovery Survey of US consumers (August 2025, n = 1,927), half said they intentionally seek out AI-powered search, and 44% of AI search users called it their primary and preferred source of insight for buying decisions, ahead of traditional search at 31% (McKinsey, 2025 (opens in a new tab)). Adobe reported that traffic from generative AI sources to US retail sites rose 693% year over year in the 2025 holiday season (Adobe (opens in a new tab)).

And the clicks that used to follow a search are thinning out. Pew Research Center found that Google users who saw an AI summary clicked a traditional result on 8% of visits, compared with 15% when no summary appeared, and clicked a link inside the summary on just 1% (Pew, 2025 (opens in a new tab)). If the answer is the destination, you need to be in the answer.

How AI assistants decide what to recommend

There's no published ranking algorithm, and each engine differs. But the broad mechanics are understood well enough to act on.

Two sources of "knowledge". Large language models carry general knowledge from training data, which can be months or years old. For current questions, most assistants also retrieve live web pages through a search index, then summarize them. ChatGPT search uses OpenAI's crawler, OAI-SearchBot, to index sites (OpenAI crawler docs (opens in a new tab)); Perplexity uses PerplexityBot (Perplexity docs (opens in a new tab)); Google's AI Overviews and AI Mode draw on Google's own index (Google Search Central (opens in a new tab)).

Many sources, mostly not yours. Answers are synthesized from several pages: review sites, comparison articles, forums, marketplaces, news, Wikipedia-style references and brand sites. McKinsey's analysis found that a brand's own sites often account for only 5 to 10 percent of the sources AI search references. Your reputation in AI answers is largely what other people publish about you.

Clear, specific, verifiable text wins. Models favour passages that directly answer the question with concrete facts. In the academic paper that coined the term, researchers found that adding citations, quotations and statistics to a source made it noticeably more visible in generated answers on their benchmark (Aggarwal et al., GEO, 2023 (opens in a new tab)). The effect varied by domain, and it was measured in a test environment, so treat it as direction rather than a promise.

Consistency builds confidence. When your name, category, prices and policies are described the same way across many sources, assistants repeat them. When they conflict, assistants hedge, pick the most common version or leave you out.

Step 1: measure your AI visibility

You can't improve what you only spot-check. AI answers vary between runs, users and days, so a single screenshot is close to meaningless. Build a repeatable panel instead.

1. Write a prompt set. 50 to 150 questions real buyers ask, grouped by intent:
   • Category: "best [product] for [use case]", "top [service] companies in Vancouver"
   • Comparison: "[you] vs [competitor]", "alternatives to [competitor]"
   • Problem: "how do I [job your product does]"
   • Policy and trust: "who has free returns on…", "fastest shipping for… in Canada"
   • Brand: "is [you] legit?", "what does [you] sell?"
2. Run it across engines. ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Copilot at minimum. Run each prompt several times, from the location your customers are in.
3. Score each answer. Were you mentioned? Recommended? In what position? What attributes were attached (price, quality, shipping, returns)? Which competitors appeared? Which sources were cited?
4. Repeat weekly. Track rates over time: mention rate, recommendation rate, share of voice against named competitors, and attribute accuracy.

Also check your analytics. Many assistants pass a referrer when users click through (for example chatgpt.com or perplexity.ai), so you can build an "AI assistants" channel group and watch visits and conversions.

Step 2: fix the technical foundations

• Let the right bots in. Check robots.txt, your CDN and any bot-protection rules. Blocking OAI-SearchBot or PerplexityBot can keep you out of those engines' search results. Note that OpenAI separates search crawling (OAI-SearchBot) from model training (GPTBot), so you can allow one and block the other.
• Be indexable on Google and Bing. Google says there are no extra technical requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible for a snippet (AI features and your website (opens in a new tab)). Register with Google Search Console and Bing Webmaster Tools (opens in a new tab).
• Put key facts in HTML text. Prices, delivery times, return windows and service areas buried in images, PDFs or scripts that need clicking to load are easy for bots to miss.
• Add structured data. Organization markup (name, logo, contact, profiles), Product and Offer markup for ecommerce, and return and shipping policy markup (MerchantReturnPolicy (opens in a new tab)). It makes facts unambiguous, even though no engine promises to use it directly.
• Consider llms.txt. The llms.txt proposal (opens in a new tab) is a plain-text summary of your site for language models. It's cheap to add, but adoption by the big engines is not confirmed, so do it last, not first.

Step 3: publish answer-first content

Look at the prompts where competitors get recommended and you don't. Usually there's a page they have that you lack: a comparison, a clear policy page, a guide that answers the exact question. Then write the page that deserves to be quoted:

• Lead with the answer. One or two sentences that directly answer the question, then the detail.
• Be specific. "Free returns within 60 days, prepaid label, refunds in 3 to 5 business days" beats "hassle-free returns".
• Show evidence. Cite sources, include data you can stand behind, quote experts, show methodology. Google's guidance on helpful, reliable, people-first content (opens in a new tab) applies just as much here.
• Cover comparisons honestly. "[You] vs [competitor]" pages are among the most-retrieved formats. Fair ones get quoted; one-sided ones get ignored or contradicted.
• Use clear structure. Descriptive headings, short paragraphs, tables for specs and prices, FAQs with direct answers.
• Keep it current. Put a "last updated" date on policy and pricing pages and actually update them. Out-of-date facts are a common cause of wrong answers.

Step 4: earn the sources AI already trusts

This is where most of the gains are, and where SEO habits help least. From your measurement panel, list the domains each engine cites for your category's prompts. Typically you'll see a mix of:

• Review platforms and marketplaces (Google Business Profile, Trustpilot, Amazon, G2, Capterra, app stores)
• "Best of" and comparison articles from publishers and niche blogs
• Community discussions (Reddit, specialist forums, Quora)
• Industry directories and association listings
• News coverage and interviews
• YouTube reviews and transcripts

Then work the list:

1. Claim and complete every profile. Same name, description, category, prices and policies everywhere.
2. Ask happy customers for reviews on the platforms engines cite, in line with each platform's rules. Respond to negative reviews with facts, especially about shipping and returns.
3. Get into the roundups. Contact authors of the "best X" articles engines quote. Offer a sample, a demo, data, or a correction if they're out of date.
4. Participate where your buyers talk. Genuine, disclosed answers in communities from people at your company. No astroturfing; it backfires and violates most platforms' rules.
5. Do real PR. Original data, expert commentary and useful tools earn citations from the kind of sites assistants trust.

Step 5: correct what AI gets wrong

When an assistant states something false ("ships only within the US", "30-day returns" when it's 60), trace it. Ask the engine for its sources where it shows them, or search for the phrase. The usual culprits are an old page of yours still live, an outdated marketplace listing, a review from years ago or a competitor's comparison page. Fix or update the source, publish the correct fact clearly on your own site with structured data, and watch your weekly panel. Corrections can take weeks to flow through as engines re-crawl.

What to expect, honestly

• No one can guarantee placement. Anyone promising "rank #1 in ChatGPT" is guessing.
• Movement is uneven. Perplexity and Google, which retrieve heavily, often react faster to new content than answers that lean on training data.
• Off-site work compounds. A strong review profile and presence in trusted roundups helps across every engine and in classic SEO too.
• Measure weekly, judge monthly. Weekly data catches changes; monthly trends tell you whether the program works.

A 30-day starter plan

1. Week 1: write the prompt set and run a baseline across engines. Check robots.txt and bot access.
2. Week 2: audit facts: pricing, shipping, returns, service area. Fix your own pages and add structured data.
3. Week 3: list the top cited sources for your category. Claim and complete profiles; contact the three most important roundup authors.
4. Week 4: publish one comparison page and one answer-first guide for prompts where competitors win. Re-run the panel and compare.

For ecommerce brands, our free AI visibility checklist turns this into a tick-list. If you'd rather have it measured and run for you, see AI visibility tracking or get a free audit.