NAMED OFFERINGS · NEW Fractional CMO from $2,500/mo. Plus the 30-day AI Adoption Sprint at $25k. See both paths
Innovative Group · Ground Truth · June 2026

Extra Muros.

Outside the walls. Why the work that wins AI search happens off the page you own.

In Rome, the most important churches were built extra muros, outside the city walls. The phrase stuck to them for centuries: St Paul Outside the Walls still carries it. I think about that line when teams ask me how to show up in AI search, because the instinct is always to fortify the thing inside the walls, the website, the page, the code you control. Last time I argued that a launch is the start of the work and that a silent break can cost you the citation an answer engine would have handed your buyer. That was the inside job. This is the part most teams have not caught up to yet: the decision about whether an AI recommends you is made mostly outside your walls, in the parts of the web you do not own.

AudienceFounders, marketers, web and growth operators
AuthorAshlesha KhondAshlesha Khond, Innovative Group
PublishedJune 29, 2026
The thesis

AI search is won outside your own walls.

AI search is won mostly off your own website. A model decides whether to recommend you from what the wider web says: the mentions, reviews, profiles, and threads it already trusts. A clean, well-built site is the price of admission. Earning presence everywhere the model looks is the game.

The moat moved off your own page, and most teams are still defending the keep.

01

The wall I know how to defend

I am an engineer before I am a marketer. Give me a page and I will check it the way I learned to check any system: one clear H1, canonicals pointing where they should, schema that parses, a footer that matches every other footer on the site. That work is real and it still matters; it is where my first Ground Truth column lived, getting the page you own clean enough for a model to read. It is also the work I find most comfortable, because it lives inside the walls, where I can see every brick and fix what is wrong with it myself.

Here is what took me longer to accept. You can make the page inside your walls perfect and still lose the answer. The model reads your site, then it reads everything else, and the everything else is doing more of the deciding. The discipline that catches a broken canonical is the same discipline you have to point outward now: at the ground you do not control, where you cannot just edit the file and fix it.

I can perfect the page I own. The answer is decided by the pages I do not.
Try thisList the assets you fully control (your site, your blog) and the ones you only influence (review profiles, your Reddit and LinkedIn presence, third-party mentions). If your AI-visibility plan only touches the first list, it is aimed at the smaller half of the problem.
02

How AI search engines decide which brands to recommend

AI search engines decide which brands to name by reading your presence across the whole web, and off-site signals carry most of the weight. Ahrefs studied 75,000 brands and found that branded web mentions correlate with AI Overview visibility at 0.664, while backlinks come in at 0.218. The three strongest factors in the study were all off-site: branded web mentions, branded anchors, and branded search volume (Ahrefs, 2025). The site you control sits lower on that list than the conversation happening about you everywhere else.

There is a reason this fits how the models work. A large language model learns a brand from text: how often your name appears, what words sit next to it, and which trusted sources keep repeating it. A mention with no link still teaches the model something. So the practice that built rankings, earning links into your site, is now a subset of a bigger job: getting your name said, correctly and often, in places the model already reads.

The model does not start with your homepage. It starts with your reputation.
Try thisSearch your own brand name plus your category in plain web search. The pages that come back, the reviews, the roundups, the forum threads, are roughly the corpus an AI has learned you from. If that set is thin or out of date, that is what the model knows.
03

Which websites AI search engines cite most

AI engines cite a small set of high-trust platforms far more than anything else, and most of them are sites you will never own. An analysis of 30 million citations across ChatGPT, Google AI Mode, Gemini, Perplexity, and AI Overviews found Reddit was the most-cited source, followed by YouTube and LinkedIn, with Wikipedia and Forbes close behind (Search Engine Land, 2026). For buyer questions specifically, review platforms like G2, Capterra, and Yelp show up again and again.

This creates a gap I watch for in every audit now. A model can recommend your company in its answer and then attach the citation to a Reddit thread or a G2 page, because that is the source it trusts to back the claim. Your competitor can win the recommendation while a third party gets the link. If your brand is missing from those rooms, you are missing from the evidence the model reaches for, and a clean website cannot make up the difference.

Your buyer's AI is reading Reddit and G2 about you. The question is whether you are in the conversation.
Try thisTake three platforms from that list that fit your business (for B2B, often Reddit, YouTube, G2). Check whether your brand has an accurate, current presence on each. A blank or stale profile is a room where your competitors are talking and you are not.
04

How to get your brand cited by ChatGPT and Perplexity

You earn citations in ChatGPT and Perplexity by building an accurate, well-supported presence in the places those engines already pull from, then making your own pages easy to quote. The off-page half is digital PR in plain clothes: get your company discussed in credible articles, claim and fill your third-party profiles, show up usefully in the communities and on the video platforms where your buyers actually ask their questions. This is the engineering behind a 90-day off-page plan, and it is slow on purpose, because trust accrues. It also takes senior, cross-channel attention, the kind of program a fractional marketing leader is built to run.

The on-page half still helps, and here the research is specific. The Princeton-led study that named generative engine optimization tested what makes a page more quotable and found that adding well-placed statistics, citations, and direct quotations lifted a page's visibility in AI answers by up to 40% (Aggarwal et al., KDD 2024). Write so a model can lift a clean, sourced sentence from your page, and you make yourself easy to cite. That is the same standard I hold this column to, which is why every claim here carries a link.

Be quotable on your own page, and be present on the pages others trust. You need both.
Try thisPick one buyer question, write a 40-to-60-word answer to it backed by a real, linked statistic, and publish it where it belongs. Then find one Reddit or industry thread the AI already cites for that question and add something genuinely useful. One inside move, one outside move.
05

How to measure your brand's visibility in AI search

You measure AI visibility by tracking how often, and how well, you get cited for the questions your buyers actually ask. Pick ten real buyer prompts. Run them across ChatGPT, Claude, Perplexity, and Google's AI mode on a schedule. Record who gets named, who gets the citation link, and which sources the model leaned on. That citation share is your scoreboard, and it tells you something a ranking report cannot: whether the web outside your walls is vouching for you.

None of this was handed to me as a task. I treat it as mine because the work does not stop at the property line of our own site, and someone has to watch the ground beyond it. The engineer in me wants a number to move and a check that runs whether or not anyone remembers to look. So this becomes a standing audit with a named owner, the same way page consistency did. If no one on your team is running that check yet, it is the kind of work we take on at IG. What I refuse to leave to chance is whether we can see ourselves the way the model sees us.

If you cannot measure your citation share, you are optimizing in the dark.
Try thisOnce a quarter, run your ten buyer prompts through the major AI engines and log who gets cited. Watching that list move, or fail to, will tell you more about your AI visibility than any single ranking.
Appendix

What I refuse to call done.

  • A brand with a perfect website and no presence anywhere the model actually reads.
  • A review or directory profile left unclaimed, blank, or two years out of date.
  • A page that buries its best fact in paragraph four, where no model will lift it.
  • An off-page plan with no owner and no schedule, so it quietly stalls after month one.
  • An AI-visibility claim with no citation-share check behind it.

Outside the walls

The churches the Romans cared most about went up extra muros, outside the safety of the walls, because that was where the ground that mattered was. AI search works the same way now. The site inside your walls has to be sound, and I will keep checking mine line by line. The recommendation, though, is won outside: in the mentions, the reviews, and the threads where a model decides who to trust. The team that does the patient work out there, long after the website looks finished, is the one the answer engine learns to name.

Next in Ground Truth: the best content your company has is already in your experts' heads. Here is how to get it out.

Frequently Asked Questions

Generative engine optimization is the practice of improving how often your brand and pages get cited in AI-generated answers from tools like ChatGPT, Perplexity, and Google's AI mode. The term comes from a Princeton-led study (KDD 2024) that tested which content changes raise a page's visibility in AI answers (Aggarwal et al., 2024).

Traditional SEO works to rank your pages in a list of links. GEO works to get your brand named and cited inside a single AI-generated answer. The two share fundamentals like clear structure and credible content, while GEO leans more heavily on off-site signals: an Ahrefs study of 75,000 brands found the strongest factors for AI visibility were all off-site (Ahrefs, 2025).

Answer engine optimization is structuring your content so an answer engine can extract and reuse it directly, usually as a concise, factual answer to a specific question. In practice it overlaps heavily with GEO and with good SEO. Innovative Group covers the practical steps in its AEO guide and on the Insights blog.

Mostly no, at least not yet. The same Ahrefs study found paid factors like branded ad traffic (0.216) and ad cost (0.215) showed only weak correlation with AI Overview visibility, far below earned signals like brand mentions (Ahrefs, 2025). AI visibility today is earned through reputation and presence across the web.