An assistant can only name what is in front of it when it answers. That is the whole mechanism, and it is why the question of how to get mentioned resolves into a question about supply: what the model absorbed in training, what it retrieves live from the web for this particular query, and what the platform has sold. Each of those is reachable by a different method, on a different timescale, with a different kind of evidence behind it. This guide sets out all three, what the current data says each one is worth, and the measurement problem that sits underneath every one of them.
How AI assistants actually choose which brands to name
When someone asks an assistant for a recommendation, the answer is assembled from two pools. There is what the model learned during training, which is fixed until the next model. And there is what it fetches live for that specific question, which is not fixed at all and is where most current, comparative and commercial answers come from. The assistant retrieves a set of pages, reads them as text, and synthesises a response that names the brands appearing credibly in that material. Present in context, and you are mentioned. Absent, and you are not. Ranking is no longer the mediating variable. What the assistant read is.
The practical complication is that each assistant reads differently, and reads a different amount. Semrush's AI Visibility Index for the first half of 2026 found ChatGPT drawing on an average of 15.4 sources per answer, weighted heavily towards Wikipedia and Reddit. Google's AI Mode averaged around 11.4 sources, leading with Facebook, Instagram and Yelp, while YouTube was the single largest source for AI Overviews. Gemini averaged just 3.3 sources, favouring YouTube, Wikipedia and e-commerce sites. A strategy that gets you named in ChatGPT is therefore not the same strategy that gets you named in Gemini, where three sources decide the answer and the bar for inclusion is far higher. Journalism is a meaningful share of the underlying supply: the 5W AI Visibility Index puts at least 27% of AI citations as coming from journalism, rising to 49% on time-sensitive queries.
How often does the average brand actually appear?
Rarely, and less often than most marketing teams assume. Webflow's AEO Maturity Index, which assessed 2,000 websites on published content, technical readiness, brand authority and measurement, found the median company appeared in 16% of relevant AI answers and appeared in a citation with a link only 6% of the time. Guy Yalif, chief evangelist at Webflow, attributed most underperformance not to exotic GEO failures but to ordinary hygiene problems: broken links, missing metadata and stale content.
Scale helps considerably. Large companies in the same dataset showed roughly double the mention rate, 60% higher citation rates and 80% higher share of voice in AI search. The self-assessment gap is wider still. Scribewise and Scrunch's 2026 AI search survey found 67% of marketers saying their brand or their client's brand appears less frequently in AI answers than they would like, while 71% were not tracking share of voice against competitors, 70% were not monitoring sentiment and 67% were not analysing AI bot traffic at all. Most brands are unhappy with a number they are not measuring.
Route one: earn the citation
Earning organic mentions is the discipline of generative engine optimisation, and the fundamentals are less exotic than the acronym suggests. Assistants favour content that is factual, clearly sourced, well structured and current. In practice that means publishing original data rather than assertion, structuring pages so a direct answer is extractable, making your brand and product entities unambiguous across the web, keeping material fresh, and building presence across the range of sources different assistants trust.
Two things are worth stating plainly because they are usually skipped. First, format matters more than novelty. As Jen Cornwell of Tinuiti has put it, a great deal of current best practice is what SEO was doing ten to fifteen years ago: an FAQ, a bullet list, an HTML table. Those are formats every LLM reads cleanly. Second, the content type with the strongest citation evidence is original research. NP Digital's May 2026 survey of 500 marketers found original research cited in AI search at a rate of 82%, far ahead of every other format, with video at 2%.
The other lever is earned media and social. Cornwell expects a larger share of LLM influence in 2026 to come from organic social, and Reddit is repeatedly named as the highest-leverage surface because its archive is treated as authority by the major models and older posts continue to be sourced. The limitation of the whole route is speed and control. It compounds, but it is slow, it cannot be aimed at a particular user moment, and you cannot choose which queries you appear in.
Route two: buy a labelled slot from the assistant
This route did not exist in a usable form a year ago. OpenAI began testing ads in ChatGPT on 9 February 2026 for logged-in adult users on the Free and Go tiers in the United States, expanded to Canada, Australia and New Zealand in the spring and then to the United Kingdom, Mexico, Brazil, Japan and South Korea, and has since added self-serve buying, product feed campaigns and a Google Shopping-style product carousel ahead of the fourth quarter.
The important detail for anyone planning a strategy around it is what OpenAI says the format does not do. Ads appear as a clearly labelled sponsored card, visually separated from the answer, and OpenAI states plainly that ads do not influence the answers ChatGPT gives. Selection is contextual, matching submitted ads against the topic of the conversation, past chats and previous ad interactions, with advertisers receiving only aggregate performance data and no access to conversations. Ads are excluded from accounts predicted to belong to under-18s and are not eligible near sensitive or regulated topics including health, mental health and politics.
So this route buys you adjacency, not inclusion. It puts a sponsored unit next to the answer. It does not put your brand inside the answer, and by design it cannot.
Route three: be present in the content the assistant retrieves
The third route works on the supply side rather than the interface. If the answer is assembled from pages the assistant fetched moments earlier, then the material in those pages is the raw input to the response, and placement inside it is placement inside the reasoning rather than beside the result.
This is the route blankspace provides for advertisers. You define specific, accurate statements about your product, called brand facts. You target by IAB content category, geographic market and a small number of prompt clusters, which are the actual questions users ask when researching your category. When a Live Search Agent retrieves a matching publisher page, blankspace injects the brand fact into the content as editorial context through a sub-50ms auction at the CDN edge, before the agent has finished reading the page. The agent reads it as part of the publisher content.
blankspace is not the only party working this layer, and the honest position is that the category is young. Time began serving ads to AI agents in July 2026 through Mobian, with Ally Bank and the Project Management Institute as first buyers, one agent ad per markdown page, sold on contextual and franchise targeting. The buy-side response has been openly sceptical. Buyers interviewed by Digiday in August questioned whether specialised agent formats are worth separate budget, with Stephan Kopp of Mediaplus Performance arguing developers will eventually route around the workaround, and Rita Steinberg of FUSE Create making the sharper point that you can influence AI visibility more reliably than you can currently buy it. A WARC study conducted by Charlie Oscar estimated 63% of brand visibility comes from long-term brand equity against 26% from current activity, which is a real argument for not treating any of this as a substitute for brand building.
Which route does which job
Use the organic route to build durable presence, because it is the only one that compounds and the only one that survives a platform changing its ad policy. Use platform ad products where you want reach against a labelled commercial slot and can accept that the answer itself is untouched. Use retrieval-layer placement where the objective is to be present in the material behind the answer for a defined set of high-intent prompts, and where you want category-level data on which questions and content categories actually drive visibility before your competitors have it.
None of the three is a substitute for the others, and the sequencing matters more than the split. Measure first, because a mention rate you have not baselined cannot be improved on paper.
The measurement problem you have to solve first
Here is the finding that reframes the whole exercise. Similarweb's study The Downstream Impact of AI Visibility, published on 21 June 2026, used opted-in US desktop clickstream data from July to December 2025 to follow real users who received a ChatGPT brand recommendation across finance, travel and beauty, excluding anyone who had named the brand in their own prompt. Recommended brands were 2.5 times more likely to receive a site visit within the following seven days than their paired competitor. That effect was symmetrical: when the competitor was recommended instead, the traffic went to them.
The consequential number is where that traffic arrived from. Only 8.8% of AI-influenced visits came through the AI channel as a trackable referral. A full 55.9% arrived via search, against 40.4% for the non-influenced cohort, because users remember the recommended brand and go and search for it by name days later. In your analytics that lands as organic branded search with no referrer, no UTM and no session chain. The AI did the persuading and Google collected the attribution. Rand Fishkin, who contributed to the study, compares it directly to the way twentieth-century advertisers measured billboards and radio through lift in store visits, and the comparison is apt: this is a lift study, not a click path.
Engagement follows the same pattern. AI-influenced visitors averaged 12.0 pages and 11.8 minutes per session against 6.5 pages and 5.6 minutes for everyone else, which is what you would expect from someone who did their comparison work in the conversation and arrived having already decided.
What this means practically is that measuring AI visibility through referral traffic alone will systematically underreport it by a large factor. Treat visibility as a leading indicator, defend your branded terms in search because that is where the demand actually lands, and benchmark against named competitors rather than tracking your own score in isolation, since the dynamic is zero-sum.
What the tools can and cannot tell you
The AI visibility tool market has grown fast and consolidated faster. Adobe completed its acquisition of Semrush for $1.9bn in April 2026, Microsoft has extended Clarity to report which pages are referenced in AI answers, and Profound, Peec AI, Otterly.AI, AthenaHQ, Evertune and Ahrefs Brand Radar all compete for the same budget. Pricing runs from roughly $99 a month at entry level to $1,000 a month or more for enterprise coverage.
Buy them with clear eyes. Paul Dyer of /prompt puts the core problem bluntly: give three different tools the same prompts and you get three different answers. Heather Physioc of VML notes that most tools still return point-in-time results rather than continuous measurement, so a genuinely trended view of discoverability across channels is not yet feasible. Ryan Mason of Markacy's verdict, that these are benchmarkers rather than sources of truth, is the right expectation to set internally. The thing they do reliably, which is worth paying for, is telling you how often you appear across a fixed prompt set relative to competitors.
What this is costing everyone else
Budget is moving, unevenly. Fractl's 2026 survey found marketers routing about 24% of search or content budgets to AI visibility work, with 82% having allocated at least something, 43% spending more than 20% and 18% allocating nothing at all. The spend is concentrated in performance marketing teams at 32% and SEO at 31%, with content at 20% and brand at 21% spending least, which tells you the discipline is currently being run as a performance problem rather than a brand one.
Execution lags the budget. 10Fold's 2026 content report found 52% of B2B marketers naming AI-generated search and answer engines as their most effective distribution channel, against 29% for organic search, while the largest group of respondents said only a quarter to a half of their content has been created or optimised for AI discovery.
Where blankspace fits
blankspace operates at the CDN edge, identifies Live Search Agent retrievals as distinct from human sessions, and runs a real-time auction to place advertiser brand facts inside the publisher content the agent is reading, with same-day attribution on the retrieval side. That is a supply-side answer to the third route above, and it does two things the other two routes do not: it puts the message in the material the answer is built from, and it gives the publisher a revenue line against traffic that otherwise monetises at nothing.
It is not a replacement for earning citations, and it does not close the attribution gap described above, because no vendor can see the branded search a user performs three days later on a different property. That limitation holds regardless of which vendor you use, and any supplier claiming otherwise is describing a measurement they do not have.
Frequently asked questions
Can you pay to appear in ChatGPT or Perplexity answers?
You cannot buy a place inside the model's own answer. OpenAI states that ads in ChatGPT do not influence the answers it gives and are shown as clearly labelled sponsored cards separated from the response. What you can buy is a labelled slot beside the answer, or placement inside the publisher content a live retrieval agent reads while assembling one.
How is a brand fact injected into publisher content different from an ad?
A brand fact is an accurate factual statement about your product placed in the editorial content an agent retrieves, so it is read as context rather than rendered as a display unit to a human. A sponsored card in ChatGPT is a labelled advertisement served alongside the answer. They occupy different positions in the pipeline and are bought and measured differently.
How do I get mentioned organically?
Publish factual, clearly sourced, well-structured content, keep it current, fix the ordinary hygiene problems first, make your brand entity unambiguous, and build presence across the sources each assistant actually draws on. Prioritise original research, which NP Digital found cited at 82%, and treat earned media and Reddit as part of the surface rather than a separate channel.
Does appearing in AI answers actually drive traffic and sales?
The best current evidence says yes, with a delay and through a channel you will misattribute. Similarweb's June 2026 clickstream study found brands recommended by ChatGPT were 2.5 times more likely to receive a site visit within seven days, and that 55.9% of that traffic arrived through branded search rather than as an AI referral. Those visitors also viewed roughly twice as many pages per session.
How do I know whether my brand is being mentioned now?
Run a fixed set of category prompts across ChatGPT, Gemini, Perplexity and Copilot and record where you appear, or use an AI visibility tool to do it continuously. Measure each assistant separately, because source counts and source preferences differ sharply between them, and treat any single tool's number as a benchmark rather than ground truth.