MiraXShow up when they ask ChatGPT.
There are exactly two ways into an AI's answer: buy your way in with ads, or earn your way in with content the model cites. Most teams are doing neither, and a few are doing the first one badly.
Half your marketing is invisible to the model.
Answer engines build their view of your category out of text they can crawl. Everything else leaves no trace, however well it performed. Worth knowing which half of your spend is which.
Read by models
Crawlable, permanent, and quotable with attribution. This is the entire raw material of an AI answer.
- Reddit and forum threads, heavily weighted and licensed at scale
- Independent comparison and review articles
- Review directories and marketplace listings
- Long-form video with a real transcript
- Documentation, changelogs, technical writing
Invisible to models
Can still be the best money you spend. Just don't expect it to change what ChatGPT says about you.
- TikTok: walled, with no reliable public transcript
- Instagram posts, reels and stories
- Paid social creative of any kind
- Anything behind a login or an app-only wall
- Your own blog, mostly — discounted as vendor copy
ChatGPT ads are prompt context, not a feed buy.
Two things come up in every conversation: nobody knows how to set these up, and the numbers look worse than Meta. Both are true, and neither is the problem people think it is.
The Meta benchmark is the wrong one
Different job. Meta interrupts someone scrolling; this reaches someone mid-decision who is actively describing their problem. Compare cost per qualified signup, never cost per impression, or you'll kill a channel that was working.
It's low volume by design
The inventory is small and always will be. Treat it as the bottom of your funnel and a source of unusually clean intent data, not as a growth engine you scale until it breaks.
The setup is the whole game
What earns placement is the context you attach to: the shape of the questions you want to appear inside, and copy that answers instead of interrupting. This is closer to writing documentation than writing an ad.
Citations are the half that compounds.
Ads stop the day you stop paying. A cited article answers a buying question every day, indefinitely, at no marginal cost. The catch is that you can't write it yourself. Not credibly.
It has to be somebody else's name
Models discount first-party marketing, and buyers do too. Value runs roughly in this order, best first:
- Genuine forum and community discussion
- Independent review sites and directories
- A creator's own newsletter, blog or long-form video
- Guest and syndicated writing
- Your own blog, worth doing but worth the least
Why the roster matters
Third-party authorship at any volume needs people willing to put their name on it. That's a creator roster, the same one running your campaigns. One more line in the brief turns a decaying video into a permanent citation.
- The companion piece is scoped into the brief, not bolted on after
- The creator keeps authorship and their own voice
- Placement is chosen for where models read, not for reach
- Planted threads get you banned and become the story
An invisible channel you can still manage.
There's no referrer on a ChatGPT recommendation. Your analytics will show direct traffic and shrug. So we instrument it two ways. Neither is perfect on its own.
Share of answer, monthly
A fixed set of 40 questions your buyers ask, run every month across ChatGPT, Perplexity and Google's AI results. We track whether you are named, in what position, and which source got you there.
- Same questions every month — the trend is the point
- Competitors tracked on the same set
- Attributed back to the specific piece that earned the mention
How did you hear about us
One field at signup, asked well. It's the only reliable way to catch discovery that leaves no click behind. We help you place it so it gets answered instead of skipped.
- Free text, so people can name something you didn't predict
- Read weekly, coded monthly
- Cross-checked against the campaign calendar
Two clocks, and we say so on day one.
The most common way this work fails is a client expecting the slow half to behave like the fast half, and pulling the plug at month two.
Baseline and ads live
Your 40 questions scored before we touch anything, and the paid lane running.
Creators live, artifacts published
Campaign spike arrives. The durable pieces go out and start getting indexed.
The answer moves
Citations show up in answers. This is where the compounding half starts paying.
The questions we get asked.
Why do ChatGPT ads perform worse than Meta or TikTok?
Usually because they're being measured against the wrong benchmark. ChatGPT ads aren't a volume channel. The impression count will always look small next to a Meta buy. What they are is intent capture at the moment of decision, with almost nobody bidding against you. Judged on cost per qualified signup rather than cost per click, the picture usually inverts. The second reason is setup: most teams port over a social-feed structure of audiences and creatives, when what actually drives performance is the prompt context you're attaching to.
Does posting on TikTok help me show up in ChatGPT?
Almost not at all, and this surprises people. TikTok and Instagram are walled off from the crawlable web: no reliable public transcripts, no indexable text. AI answer engines build their picture of your category from text they can read: comparison articles, review sites, forum and Reddit threads, long-form video transcripts, documentation, news. A viral TikTok can be enormously valuable and still leave zero trace on the answer surface.
What is generative engine optimization?
Getting named when someone asks an AI a buying question. The mechanics are closer to PR and community than to old-school SEO: what matters is being described favorably by third parties in places models read, not keyword density on your own pages. Content on your own blog is the weakest form of it — models discount obvious vendor material.
How do you measure something with no click and no referrer?
Two instruments. First, a fixed set of 40 real buyer questions, run monthly across ChatGPT, Perplexity and Google's AI results, tracking whether you're named and in what position — a share-of-answer report. Second, a how-did-you-hear-about-us field at signup, which is the only honest way to catch dark-social discovery. Neither is perfect. Together they turn an invisible channel into something you can manage.
How long does generative engine optimization take to work?
Ads work immediately, because you're buying placement. Citations take months, because content has to be published, indexed, and then picked up by models on their own schedule. Anyone promising to change what ChatGPT says about you in three weeks is selling something. We pair the two deliberately so there's something moving while the slow half compounds.
Find out where you stand.
The first thing we do is score you on the forty questions your buyers are already asking. Most founders have never seen that number.