How do you get an AI assistant to recommend your brand?
You get recommended by becoming the consensus answer in the sources assistants read: category listicles, community threads, review platforms, and comparison pages, backed by a site structured so machines can quote it. There is no ad placement to buy. You earn the recommendation off-site, then verify it monthly with a prompt panel across ChatGPT, Perplexity, and Gemini.
TL;DR
- Assistants recommend by consensus: the brands named consistently across listicles, Reddit threads, review platforms, and comparison pages win the answer. Nobody can buy the slot.
- Map the prompt universe first: the 30 to 50 phrasings your buyers actually type, from "best X for Y" to problem descriptions.
- Branded mentions correlate with citations more strongly than backlinks, per Ahrefs' study of 25 million AI citations. Placements beat link building.
- Your site has one job in this system: be extractable. Answer-first pages, schema markup, open AI crawler access, refreshed before the ten-month freshness window closes.
- Track it like a channel: a monthly prompt panel in clean sessions, logged per query, with the cited sources doubling as your outreach list.
- Honest timelines: weeks for retrieval surfaces to see a fresh page, quarters for the recommendation itself to move.
Assistants pick brands on consensus, not ads
Ask ChatGPT for the best magnesium for sleep and watch what comes back: three or four named brands, a sentence on each, sometimes product cards with prices. That answer was not ranked. It was assembled, from what the model already believes about the category plus whatever it retrieved in the moment: roundups, review platforms, community threads, comparison pages. The brands that appear are the ones those sources keep naming, described the same way, over and over. Consensus is the selection system.
Notice what is missing from that process: a bid. There is no auction behind the answer, no placement to purchase, and even the shopping cards ChatGPT drops into a buying conversation are filled from organic signals, not media budgets, today. That cuts both ways. You cannot pay your way in this quarter, and a competitor cannot outbid you out of the answer once you hold it.
One more mechanic matters for everything that follows. Assistants split a broad ask into narrower sub-queries before they answer, so "best magnesium for sleep" quietly becomes questions about absorption, dosage forms, and brands that work for sensitive stomachs. Presence across those narrow questions pulls you into answers the broad phrase alone would never earn you.
We broke down the three-layer model behind all of this in our AEO and GEO explainer. That article is the what. This one is the how, in the order we run it.
Map the prompt universe before you build anything
The prompt universe is the complete set of questions your buyers put to assistants, written in their words, not yours. It comes first because every later decision (which placements to chase, which pages to write, what to track) reads directly off this list.
Mine four places for language. Support tickets and post-purchase surveys tell you how buyers phrase the problem. Search query reports in your ads accounts show the comparisons they already run. Autocomplete on Google and inside ChatGPT surfaces phrasings you would never invent at a whiteboard. And your category's community threads hand you the exact vocabulary, complaints included.
Sort what you find into four intent buckets: category asks ("best sulfate-free shampoo for colored hair"), comparisons ("brand X vs brand Y for fine hair"), problem phrasings ("hair breakage after bleaching, what helps"), and use cases ("safe options to use after a keratin treatment"). For most DTC brands the list settles between 30 and 50 prompts. Short enough to run by hand, wide enough to cover the fan-out.
Selling across Europe? Build the panel per language. Assistants answer German buyers from German sources, and we have watched brands own the default answer in English while being absent from the same question asked in their second-biggest market. One panel per market, or you only see half the scoreboard.
The list does double duty from day one. It is the tracking panel you will run monthly, and it is the build roadmap: every prompt either names your brand in the answer or exposes the missing placement or missing page that explains why not.
Win the sources assistants actually read
Most brands get this exactly backwards: months polishing their own site while the assistant assembles its answer from other people's pages. Ahrefs' study of 25 million AI citations found that branded mentions correlate with citations more strongly than backlinks. Being named in the roundup now matters more than being linked by it, which is not how most SEO budgets are currently pointed.
Four source types do most of the work:
| Source | What the assistant takes from it | Your move |
|---|---|---|
| Listicles and category roundups | The candidate list for best-of answers | Pitch inclusion in the roundups your prompt panel shows getting cited; fresh articles stay retrieved, stale ones age out |
| Community threads (Reddit, forums) | Unfiltered consensus and real usage stories | Participate honestly where your category is discussed; shill posts get removed, genuine answers keep resurfacing for years |
| Review platforms | Volume, recency, and the words customers use | Keep recent review volume flowing; recency reads as relevance to retrieval systems |
| Comparison pages | Head-to-head positioning language | Publish honest vs pages on your own domain before a third party frames the matchup for you |
Two rules govern the outreach. Target provably retrieved sources first: a page your prompt panel shows being cited is worth more than ten generic PR placements, because you know the assistant reads it. And enforce one description everywhere: same brand name, same one-line positioning, on every placement, profile, and seller page. Assistants build confidence through repetition. Five slightly different descriptions of the same brand read as noise, not consensus.
Make your site something a machine can quote
Off-site mentions win you the nomination. Your site still has to survive the fact-check, because retrieval surfaces open your pages to verify what the roundups claim, and what they can parse decides whether you get quoted or silently skipped.
The fix order we run:
- Open the doors. robots.txt must allow GPTBot, PerplexityBot, and ClaudeBot. A blocked crawler makes every other move on this list irrelevant, and we still find blocks on stores actively paying for AI visibility work.
- Lead with the answer. Every page mapped to a prompt opens with a direct answer to that prompt, details after. Machines quote what sits at the top.
- Mark it up. Organization, Product, and FAQ schema, so the assistant reads structure instead of guessing at prose.
- Cover the long tail. One narrow question per page, mirroring the fan-out. The narrow page gets retrieved into answers your homepage never reaches.
- Stay fresh. Roughly 95% of AI citations in the same Ahrefs study pointed at content under 10 months old. A refresh cadence is not polish; it is the entry fee for staying in the pool.
None of this is exotic. It is the same answer-first discipline we detail in our breakdown of the three AI search layers, applied page by page against your prompt list.
Track visibility like a channel, not a curiosity
You cannot manage a channel you never measure, and AI recommendations are now a channel. The tracking discipline is refreshingly manual. Once a month, run every prompt in your universe through ChatGPT, Perplexity, and Gemini in fresh, logged-out sessions so personalization does not flatter you, and log three columns per query: named or absent, which competitors appeared, and which sources the answer cited. A spreadsheet is enough. Screenshot the answers as you go: assistants change their minds between runs, and the month-over-month trail is the only way to separate a real trend from answer noise.
That third column is the quiet goldmine. Cited sources are a self-updating map of the pages worth winning a mention on, and when a competitor gets named while you do not, the citations show which placements earned them the slot. Your outreach list writes itself, one panel run at a time.
Downstream, watch the numbers that confirm the loop is working: branded search impressions climbing and blended acquisition cost drifting down, both defined in our DTC metrics glossary. They move on a lag. The prompt panel is the leading indicator, and it shows movement months before the P&L does.
Expect weeks for retrieval, quarters for recommendations
Anyone quoting you a 30-day ChatGPT ranking has confused a screenshot with a position. The honest arc has two speeds. A fresh, well-structured page can enter retrieval-based answers within weeks, because surfaces like Perplexity read the live web on every query. The recommendation itself, where an assistant names your brand unprompted in a category ask, moves over quarters, because it follows the slow accumulation of mentions across independent sources. Budget it the way you budget infrastructure, not a promo: a monthly operating rhythm, reviewed quarterly, judged on share of answers rather than next week's revenue.
The compensation for the wait is stickiness. Once you are the consensus answer, a competitor faces the same slow climb you did, this time against an incumbent. In most DTC categories that position is still winnable, which is the entire argument for starting before your category wakes up.
Start with the audit, not the retainer
This playbook is the operating manual behind our AEO and GEO practice, run monthly: the prompt panel calibrated to your category, the mention-gap map against the competitors already being named, and the content and placement work sequenced by what the panel keeps surfacing. We publish the method because the method is not the moat. Running it every month without getting bored is.
If AI search is anywhere on your roadmap, do the cheap step first: a visibility audit through our AEO and GEO service for ecommerce brands that shows which answers name you today, which ones a competitor already owns, and the shortest route between the two. The brands being cited right now are turning into their categories' default answers, and defaults get harder to displace every month. Find out where you stand while changing the answer is still cheap.
