Ranking And Being Cited Are Now Two Different Games.
A 40,000-keyword study found 88% of AI Mode citations do not come from the organic top ten. Analysis of 75,000 brands found web mentions correlate with AI citation roughly three times more strongly than backlinks. Your rank tracker is measuring a shrinking share of reality.
What is an AI marketing agency?
An AI marketing agency, sometimes called an artificial intelligence marketing agency, does two distinct things, and it is worth separating them. The first is making your business visible inside AI systems — being named when someone asks an assistant for a recommendation. The second is using AI inside your marketing operation to produce work and handle tasks faster than a team could alone. Most firms calling themselves an AI digital marketing agency do only the second and describe it as the first.
We will be direct about something, because the category is full of noise: a great deal of what gets sold as AI marketing is ordinary marketing with a tool in the workflow. Using a language model to draft copy faster is useful. It is not a strategy, and it does not make a firm an AI marketing company or an AI advertising agency any more than owning a camera makes someone a film studio.
What has genuinely changed is where buyers now ask their questions. When somebody describes a problem to an assistant and gets three companies named back, the ranking that used to decide who they found has been replaced by a citation decision made inside a model. That decision runs on different signals than search rankings do, and the published research on those signals is where this page spends most of its time.
Watch an assistant
decide who to name
Five businesses could reasonably answer this question. Only some get named. The signals below are weighted according to published 2026 correlation research — brand mentions across the web, structured direct answers, third-party presence, freshness and entity strength. Toggle your own signals at the bottom and run it again.
Signal weightings reflect published 2026 correlation research, including Ahrefs' analysis of 75,000 brands (brand web mentions 0.664 correlation with AI citation against 0.218 for backlinks, and video mentions the single strongest factor at 0.737), The Digital Bloom's LLM ranking factor work (brand search volume 0.334; branded entities roughly 6.5 times more likely to be cited), and SE Ranking's 129,000-domain study on community mentions. Correlation is not causation, and no external party can see the actual retrieval logic. This is a teaching model of published relationships, not a simulation of any specific engine.
The uncomfortable part
Four of the five signals being scored are not on your website. They are mentions in other people's content, presence on third-party lists and reviews, video, and how often people search your name directly. That is a substantially harder problem than publishing a page, and it is why agencies prefer to sell you the page.
It also explains a pattern that confuses a lot of businesses: a site can rank respectably and be almost entirely absent from AI answers, while a competitor with worse rankings gets named constantly because they are talked about everywhere else.
One strategy cannot
cover both engines
Analysis of roughly 680 million citations found that only 11% of domains cited by ChatGPT are also cited by Perplexity. These are not two versions of the same thing. They retrieve differently, weight differently, and reward different work.
The gap in how often they name brands at all is even wider. A study of 34,234 AI responses reported ChatGPT citing brands 0.59% of the time, Perplexity 13.05%, and Grok around 27% — a spread of roughly 46 times.
Brand citation rates from a 2026 analysis of 34,234 AI responses; domain overlap from an analysis of approximately 680 million citations across ChatGPT, Google AI Mode and Perplexity. Figures move as platforms change retrieval, so treat them as a snapshot of a fast-moving picture rather than fixed values.
Why they behave so differently
| Engine type | How it answers | What earns a mention |
|---|---|---|
| Training-led | Answers substantially from knowledge already absorbed during training, retrieving from the web selectively rather than for every query. | Long-run entity strength. Being written about consistently across authoritative sources over time. |
| Citation-led | Performs live retrieval for essentially every query, reads candidate pages and synthesises with inline references. | Freshness, clear structure, and pages that answer the question directly enough to quote. |
| Search-integrated | Draws on an existing search index enriched with additional ranking for AI answers. | Conventional authority plus direct answers near the top of each section, with consistent structured data. |
The practical consequence is that work aimed at one engine frequently does nothing for another. Publishing a beautifully structured FAQ helps with live-retrieval engines and does very little for training-led ones, where the lever is being mentioned in places that get absorbed over months. Anyone promising a single tactic that wins everywhere has not looked at the overlap data.
The execution side of this — the structural, technical and content work that earns citations — runs through Eye To Ad Media, which specializes in search and AI-search optimization. What we handle here is the strategy around it and the AI work inside your marketing operation.
Test your own site
with real tools
These are not ours and we get nothing if you use them. They are the free, first-party tools we run before any engagement, and you can run all four in about ten minutes. If your site fails these, no amount of strategy matters yet.
Can a crawler render your page quickly enough to bother? Retrieval-based engines fetch and read pages live, and slow or render-blocked pages get skipped. This also happens to be one of the largest conversion factors on your site.
Does your structured data actually parse? Broken or missing schema is the most common technical reason a page that answers a question well never gets treated as an answer.
The stricter check. Google's tool only validates what Google uses; this one validates against the full vocabulary, which matters when engines other than Google are reading you.
Visit yoursite.com/robots.txt. If AI crawlers are disallowed, you have opted out of the entire channel, frequently without knowing. Plenty of sites blocked them in 2023 and never revisited the decision.
Four external links, none affiliated with us. The fastest honest test of all is simpler still: open ChatGPT, Perplexity and Google AI Mode, ask each the question your best customer would ask, and see whether you appear. Do it for three competitors too. That takes fifteen minutes and tells you more than most paid audits.
What AI actually does well,
and what is theatre
Separate from visibility, there is the question every artificial intelligence marketing agency should answer plainly: where does using AI inside your marketing actually pay? Here is our honest split after building this on real accounts rather than in a pitch deck.
| Task | Verdict |
|---|---|
| Instant lead response | Genuinely transformative. Research on response time is unambiguous, and no human team covers 9pm on a Saturday. This is the highest-return AI application in most businesses. See lead generation. |
| Creative volume | Real, with supervision. Producing enough concepts to test properly is a genuine constraint, and AI removes it. Reporting suggests AI-assisted creative can lift click-through meaningfully, largely through iteration speed. See Meta ads. |
| Qualification and routing | Works well. Asking the right follow-up questions and getting a lead to the right person is a bounded task with clear success criteria, which is where these systems are strongest. |
| Research and analysis | Useful, verify everything. Excellent at surfacing and summarizing, unreliable on specifics. Every figure on this page was checked against a named source for exactly this reason. |
| Bid management | Already happening. The platforms have automated this whether you asked or not. The skill is now supplying clean conversion signals rather than adjusting bids. |
| Unsupervised content publishing | Theatre, and risky. Volume without editorial judgment produces pages nobody wants and, given how heavily AI systems weight quality signals, can actively cost you. |
| "AI-powered" dashboards | Usually theatre. A chart is a chart. If a tool cannot explain what decision it changed, the label is decoration. |
| Strategy | Not yet. Deciding what to say, to whom, and whether the offer is any good remains judgment work. Anyone selling AI strategy is selling you a plausible-sounding average. |
As an AI advertising agency and marketing company we build the first three for clients: instant response systems and qualification agents (see AI sales agents), and creative pipelines. We are skeptical of the rest, and we would rather say so than sell you a dashboard.
Free AI visibility check
We ask five assistants the questions your customers ask, record whether you appear, and compare you against three competitors. You get the transcript either way.
Get my visibility checkWhere we start,
and in what order
- Measure before advising. We query multiple assistants with the questions your buyers actually ask and record what comes back — for you and for competitors. Almost every engagement starts with a finding nobody expected.
- Check you have not opted out. Robots.txt, render-blocking and structured data. A surprising number of sites are invisible to AI systems for reasons that take an afternoon to fix.
- Read the platforms separately. Aggregate visibility scores hide everything that matters given roughly 11% overlap. Per-engine reporting or the number means nothing.
- Fix the instant-response layer. Before anything about visibility, make sure inquiries you already receive get answered immediately. It is faster, cheaper and compounds with everything downstream.
- Build mention surface, not just pages. Since brand mentions correlate far more strongly than backlinks, the work is being talked about in third-party content, lists, reviews, video and communities — not only publishing more of your own.
- Structure what you already have. Direct answers near the top of each section, consistent schema, clear headings. This is the part that helps live-retrieval engines most, and it runs through Eye To Ad Media.
- Deploy AI where it pays. Response, qualification and creative volume. Not dashboards.
- Re-measure monthly. This field moves faster than anything else in marketing. A tactic that worked in March may do nothing by September, and the only defense is measuring rather than believing.
Where AI visibility
matters most right now
It matters most where buyers research before choosing, and least where the decision is impulsive or purely local and immediate.
Buyers research heavily before contacting anyone. Being named early shapes the shortlist.
💻SaaSComparison queries are constant and assistants answer them directly. High stakes here.
⚖Law FirmsPeople describe a problem before knowing what to search. Assistants answer that well.
🏥HealthcareHeavily researched and heavily scrutinised. Accuracy and authority signals matter more here.
🛒EcommerceProduct comparison queries are growing fast. Feed and review presence feed the answers.
🏠Real EstateLocal intent limits assistant use, but agent and area research is increasingly done this way.
🦷DentalReviews and third-party presence carry most of the weight for local practices.
🍽RestaurantsLeast affected. Decisions are immediate, local and rarely researched through an assistant.
AI marketing
questions, answered
Two separable things. The first is making a business visible inside AI systems, so it gets named when someone asks an assistant for a recommendation. The second is deploying AI inside the marketing operation — instant lead response, qualification agents, creative production at volume. Most agencies claiming the label do only the second. Be direct in asking which one is on offer, because a language model in the copywriting workflow is useful but it is not a strategy.
Yes, though the two overlap. A 40,000-keyword study found 88% of AI Mode citations do not come from the organic top ten, and separate analysis suggests around 90% of ChatGPT citations come from URLs ranked beyond position 21 in Google. Ranking well helps and is no longer sufficient. The larger difference is what the signals are: analysis of 75,000 brands found brand web mentions correlate with AI citation at roughly 0.664 against 0.218 for backlinks, so being mentioned matters considerably more than being linked to.
Because different things are being measured. Training-led systems answer substantially from knowledge absorbed over time rather than from live retrieval, so what matters is how often and how authoritatively a brand has been written about across the web — news, communities, review sites, video, reference sources. A competitor mentioned constantly in third-party content can be named far more often than a better-ranking site nobody discusses. Research also indicates branded entities are roughly 6.5 times more likely to be cited than non-branded ones.
Partially, and less than you would hope. Analysis of roughly 680 million citations found only 11% of domains cited by ChatGPT are also cited by Perplexity. They retrieve differently: one leans on absorbed knowledge and entity strength, the other performs live retrieval on nearly every query and rewards freshness and clear structure. Some work helps both — being genuinely well known and publishing clear, structured answers — but a single tactic that wins everywhere does not exist, and anyone claiming otherwise has not looked at the overlap data.
Manually, and it takes fifteen minutes. Open ChatGPT, Perplexity and Google AI Mode, ask each the question your best customer would ask before they knew your name, and see whether you appear. Repeat for three competitors. That is a more honest starting point than any tool. Paid platforms exist for ongoing tracking across engines, and they are worth it once you are actively working on this, but only if you read the per-platform breakdown rather than an aggregate score, which averages away the thing that matters.
Unsupervised AI content published at volume is bad for it, yes — not because of how it was produced but because of what it usually is. AI systems weight quality, accuracy and authority signals heavily, and pages that exist to fill a keyword slot do not accumulate those signals. AI used to draft, structure and accelerate work that a person then verifies and improves is a different activity with a different outcome. The distinguishing question is not whether a machine was involved but whether anyone with judgment reviewed the result.
Three places, in our experience. Instant lead response, because research on response time is unambiguous and no human team covers evenings and weekends. Creative volume, because testing enough concepts is a real constraint and AI removes it. And qualification and routing, because those are bounded tasks with clear success criteria, which is where these systems are strongest. We are skeptical of AI dashboards and of AI strategy, and we would rather say so.
Possibly, and it is worth thirty seconds to check. Visit yoursite.com/robots.txt and look for disallow rules aimed at AI user agents. A great many sites blocked them in 2023 when the question first arose and never revisited the decision, which means they have opted out of the channel entirely without anyone noticing. There are legitimate reasons to block crawlers, but it should be a decision rather than an inheritance.
Free AI visibility check
We ask the assistants what your customers ask, compare you against three competitors, and send you the transcript. One business day.
Your information is never sold or shared. We respond within one business day.
Related pages
Paid and owned channels only pay off when the page they land on converts. These are the pages that usually come next.
- Google Ads management — Google Ads management from a PPC agency that fixes account structure before bidding.
- Facebook ads management — Facebook ads agency built around creative production volume rather than audience settings.
- Website conversion rate — Website conversion rate explained: how to calculate it, what a good rate looks like by industry, and what one more point is worth in customers.