What generative engine optimization is
Generative engine optimization (GEO) is the work of getting a brand mentioned, recommended and cited in answers written by AI tools such as ChatGPT, Perplexity, Gemini and Google AI Overviews. You will also hear it called LLM SEO, AI search optimization or simply AI visibility. The labels differ, but the question behind them is the same: when a buyer asks an AI assistant what to use, does your brand come up?
Many of these tools search the web while they compose an answer, then list the pages they used as citations. That part is observable. Ask one for the best accounting software for freelancers, for example, and you can see which listicles, comparison posts and forum threads the answer leaned on. GEO, as I practice it, starts from those citations and works backward.
On-site and off-site GEO: the part I handle
GEO has two halves, and it helps to agree at the start who owns each one.
- On-site GEO covers your own pages: a clear structure, direct answers to the questions buyers ask, accurate product details, schema markup and making sure AI crawlers can reach and read your content. That is your team’s job, or your SEO’s.
- Off-site GEO covers what the rest of the web says about you: whether you appear in the roundups, comparisons and niche articles AI answers draw on, and how those pages describe you. That is the part I do.
I don’t edit your site, write your schema or change your crawler settings. I earn mentions for you on other people’s sites, using the same manual approach as my outreach link building. If your own pages are hard for AI crawlers to read, fix that alongside the outreach, since the two halves work best together.
The sources AI answers tend to draw on
Nobody outside the companies that build these tools knows exactly how they choose sources, and the systems change often. What anyone can do is read the citations. Run your buyers’ questions through the tools and the same kinds of pages tend to appear:
- Listicles and best-of roundups. “Best X for Y” articles on niche blogs and publications, often the same pages that rank for that query in regular search.
- Comparison and review articles. “X vs. Y” posts, alternatives lists and hands-on reviews from independent sites.
- Niche publications. Trade blogs, specialist magazines and hobby sites whose readers are exactly the people you sell to.
- Community discussions. Forum threads and question-and-answer sites where real users compare notes.
The first three are editorial. A person decides what goes in them, and that person can be pitched, which is where outreach does its work.
Community discussions are different. I don’t post in forums or seed threads for clients, because planted posts break most communities’ rules and tend to backfire once spotted. A real presence there comes from your own team taking part openly.
How a GEO outreach campaign runs
A GEO campaign runs on the same monthly plans and approval sheet as the rest of my work. What changes is where the prospect list comes from. Instead of starting only from competitor backlinks, it starts from the questions your buyers ask.
- List the prompts. With your input, I build a set of questions a buyer might put to an AI tool: category questions, head-to-head comparisons, requests for alternatives and problem-first questions where your product is one possible answer. For example, a meal kit company might track “best meal kits for families” and “[brand] vs. [competitor]” side by side.
- Check the citations. I run each prompt in ChatGPT, Perplexity, Gemini and Google AI Overviews and record which pages each answer cites and which brands it names, your competitors included.
- Sort the targets. Cited pages that are editorial and open to pitching go into the sheet with their metrics. Pages nobody can pitch, such as a competitor’s own blog, are noted and set aside.
- Pitch an inclusion or an update. For each target, I give the editor a reason to add you: a product their list doesn’t cover yet, a use case it misses or details that have gone out of date. An entry added to an existing roundup is, in practice, a link insertion, and some sites would rather publish a fresh guest article.
- Widen the net. Citations shift from one run to the next, so I also pitch similar pages that rank for the same questions but weren’t cited that day. A page that ranks for a question now may be the one an answer cites next month.
- Track and repeat. The same prompt set is checked again as the campaign runs, and the results sit in the sheet next to the placements that went live.
You approve every site before anything is published, exactly as in any other campaign, and every opportunity states its fee and link attribute up front.
How I measure AI visibility
The core measure is simple. Out of a fixed set of prompts, how many answers mention your brand, and how many cite a page that mentions you? I check this by hand across ChatGPT, Perplexity, Gemini and Google AI Overviews, once before outreach starts and again as it runs, so each round compares like with like.
Manual checks are slower than automated trackers, but they show what a buyer actually sees, including how the answer describes you. A mention that gets your pricing wrong or files you under the wrong category is worth knowing about, and sometimes it traces back to one outdated article that an editor can correct.
Alongside the prompt checks, these signals are worth watching:
- Share of answers that mention you, for each group of prompts, tracked over time.
- LLM citations: which pages the answers cite, and whether pages that mention you are among them.
- AI referral visits in your analytics, if your setup separates them from other referral traffic.
- The placements themselves, each logged in the sheet with its live URL, anchor and attribute.
The honest limits of generative engine optimization
AI answers are less stable than most people expect. The same prompt can name different brands on different runs, for different users, in different locations and after each model update. A single screenshot proves very little either way, which is why I track a set of prompts over time.
Timelines vary too. Answers that search the web can reflect a new mention once the page has been crawled again, while what a model learned in training changes only when the model itself is updated.
So I don’t guarantee citations. What I guarantee is the plan minimum: at least 20 qualified publication opportunities a month on Core and 30 on Scale, aimed at the pages and publications that answers in your category draw on. Generative engine optimization improves the odds that you are part of the answer. It can’t lock the answer in place.
The placements are real editorial content on sites with real audiences, too. They still work as backlinks and can still send readers, whatever a model does with its sources next quarter.
GEO vs. SEO: why the same links do both jobs
GEO doesn’t replace SEO. The pages AI answers cite often overlap with the pages that rank in search for the same question, and links are still one of the signals that decide who ranks. A brand that earns a place in the main roundups for its category is working on both fronts at once.
That is why I treat an editorial mention as one placement with two jobs. A linked entry in a “best of” list is a backlink for your rankings and a line of text an AI answer can quote. Every opportunity states its attribute (dofollow, nofollow or sponsored) before you approve it, so you can weigh the SEO value alongside the GEO value.
If you want steady authority first, start with outreach link building and add GEO targets to the same sheet. For stores whose buyers ask AI tools for product picks, ecommerce link building goes after the gift guides and roundups those answers often pull from. Plans and minimums for both are on the pricing page.