Why AI Answers Cite Everyone But You, and What Actually Fixes It

By Cody Wise, founder of Wise Media, Calgary. Last updated 5 November 2026.

Summary

• AI assistants answer from pages they can crawl and rank. Your own site is one small input competing with every review site, forum, directory and publication in your category.
• Google states plainly that there is no special AI markup, no llms.txt requirement and no secret schema that gets you into AI Overviews. Eligibility is ordinary Search eligibility.
• The lever most Canadian businesses have never pulled is off-site: third-party reviews, genuine community participation, and being the named source in someone else’s article.
• Mention farming does not work and Google says so directly. Earned is the operative word.
• A realistic first 90 days costs between $0 and about $4,000 CAD in outside spend, and most of it is time, not money.

Table of Contents

• Why do AI answers cite everyone but you?
• How AI engines actually pick their sources
• The owned media ceiling, and why publishing more does not break it
• The four off-site surfaces that get cited
• A 90 day plan to get cited by AI search
• What this costs in Canada
• Common mistakes
• FAQ

Why do AI answers cite everyone but you?

Because an AI answer is assembled from the pages the system can already find, rank and trust for that question, and in almost every commercial category those pages belong to someone else. A review platform. A Reddit thread. A trade publication. A government page. Your website is one candidate in a field of thousands, and it is the only one in that field with an obvious commercial interest in the answer.

That is the uncomfortable part. Most agencies selling answer engine optimisation will tell you to restructure your H2s and add FAQ schema. Those things help your own pages become quotable, and we have written the on-site version of this in detail in our piece on AI citations versus backlinks. But if you have done all of that and the answers still name your competitors, the problem is not your markup. It is that you have no presence on the surfaces the model reaches for first.

There is also a traffic reality behind this. Pew Research Center, studying the browsing behaviour of 900 US adults who agreed to share their activity, found that when a Google AI summary appeared, users clicked a traditional search result on 8% of visits, against 15% of visits where no summary appeared. Sessions also ended more often: 26% of pages with an AI summary ended the browsing session, against 16% without one. That data is American and from March 2025, so treat it as directional rather than a Canadian benchmark. The direction is clear enough. Being the cited source inside the answer matters more every quarter, and the click you used to win is not coming back on its own.

The short answer, in 50 words

To get cited by AI search, be findable on the third-party pages that already rank for your category: review platforms with real reviews, communities where buyers ask questions, and publications that quote named operators. Then make your own site technically eligible and genuinely original. Off-site presence is the gap, not markup.

How AI engines actually pick their sources

Three things are worth knowing, and all three come from the companies themselves rather than from an agency blog.

Google: AI features run on ordinary Search eligibility

Google’s own documentation on AI features and your website states that for a page to appear in AI Overviews or AI Mode it must be indexed and eligible to be shown in Google Search with a snippet, and that there are no additional technical requirements. The same documentation describes query fan-out, where a single question triggers several related searches so the system can surface a wider and more diverse set of links than classic search would.

Google’s guide to optimising for generative AI features goes further and lists what does not help. It says you do not need llms.txt, AI text files, special markup or Markdown, because Google Search does not use them. It says content chunking is not required and there is no ideal page length. It says rewriting content purely for AI phrasing is unnecessary because its systems understand synonyms and meaning. And it says, in as many words, that seeking inauthentic mentions is not as helpful as it seems and that its systems are designed to block spam.

Read that last line twice before you buy a mention package.

OpenAI: a separate crawler controls whether you exist in ChatGPT search

OpenAI publishes an overview of its crawlers that is genuinely useful and almost nobody reads. Four bots, four different jobs:

The practical consequence: plenty of Canadian businesses blocked every AI bot in 2025 on principle or because a security plugin suggested it, and in doing so quietly removed themselves from ChatGPT search while leaving model training untouched. If you want model training blocked but ChatGPT search visibility kept, block GPTBot and allow OAI-SearchBot. Check your robots.txt today rather than assuming.

The pattern across all of them

Every one of these systems retrieves from a live index of public pages and then summarises. None of them has a private channel you can buy into organically. So the question “how do I get cited” collapses into a simpler one: which public pages about my category does the retrieval layer already like, and am I on them?

The owned media ceiling, and why publishing more does not break it

Here is the thing most agencies will not put in writing, including agencies that sell content. Owned media is the weakest of the channels available for AI citation, and the reason is structural rather than qualitative.

When somebody asks an assistant “who is the best web design agency in Calgary”, the system is answering a comparison question. A comparison question is answered badly by a page from one of the things being compared. Retrieval systems, like human researchers, prefer a source that covers several options to a source that covers one. That is not a bias you can write your way out of. A self-description is still a self-description no matter how well structured.

This is why the brands that appear constantly in AI answers are rarely the ones publishing the most. They are the ones that have been written about. The mechanism is the same one that made PR valuable long before any of this existed, and the research circulating in the SEO industry through 2026 keeps landing on the same correlation: brand mentions across third-party sources track AI citation far more closely than raw backlink counts do. Those correlation figures come from vendor studies rather than from the AI companies, so treat the exact numbers with suspicion and the direction as sound.

What owned content is still genuinely for

• Google rankings. Still the largest source of qualified traffic for most Canadian service businesses and still won by depth and structure.
• Conversion. An AI answer can name you. It cannot sell for you. The page the buyer lands on does that.
• Being quotable once you are already in the consideration set. When a model is assembling an answer about your specific company, your own pages carry real weight.
• Topical depth that earns the mention. Nobody cites an operator with nothing to say. The content is the reason a journalist or a forum answers with your name.

So the correct conclusion is not “stop publishing”. It is “stop expecting publishing alone to fix an off-site problem”. Our post on internal linking for topical authority covers how to make the content you do publish behave like a cluster rather than 200 unrelated pages, which is the half of this that is inside your control.

The four off-site surfaces that get cited

In rough order of effort to payoff for a Canadian service business.

1. Review platforms, which almost everyone underuses

Review platforms rank for exactly the comparison queries that AI answers love, and they are the single cheapest off-site asset available. For Canadian agencies and B2B service providers the relevant set is usually Google Business Profile, Clutch, G2 where software is involved, and the vertical directory that owns your category.

Two rules make this work. First, the reviews have to be real and solicited within the platform’s policy. Google’s review rules are specific and enforcement has tightened, which we covered in detail in the 2026 Google review policy guide for Canadian businesses. Second, the review text matters as much as the star count. A review that says “great to work with” gives a retrieval system nothing. A review that names the service, the city and the outcome gives it a sentence it can quote. You cannot write the review, but you can ask a specific question when you request one.

An honest note, since this piece is about telling the truth: Wise Media has zero Clutch reviews as of today. It is the highest leverage unticked box on our own list, and if you are reading this as an agency owner, yours is probably the same.

2. Communities, where the phrasing matches the prompt

Reddit threads, Stack Exchange answers, industry Slack and Discord archives that are publicly indexed, and niche forums rank heavily for conversational questions because they were written as conversational questions. The phrasing a person types into an assistant is almost identical to the phrasing they type into a forum search. That is not a coincidence, and it is the reason community presence punches above its weight.

What works: answering questions in your actual area of competence, under a real identity, without a link in the first sentence. What fails: anything that reads as marketing. Communities are hostile to promotional writing, and the moderators are faster than you are.

A reasonable cadence for an owner-operator is two or three substantive answers a week in one or two communities. Not ten. Depth in one place beats presence in six.

3. Earned mentions in publications and other people’s content

This is traditional digital PR with a new payoff. Trade publications, local business press, podcasts with transcripts, YouTube videos with descriptions, and other operators’ blog posts all produce pages that rank, and your name sits inside them as a named source.

The route that works for small Canadian firms is not a PR retainer. It is being specific and available:

1. Pick two or three narrow claims you can defend with your own data or direct experience.
2. Answer journalist and creator source requests in those lanes only.
3. Go on small podcasts in your vertical, and check afterwards that the episode page publishes a transcript, because a transcript is a crawlable page and an audio file is not.
4. Give other operators something quotable when they write about your category, and let them keep the link if they want it.

One genuine trade mention is worth more than fifty syndicated press releases, and the syndicated release is exactly the inauthentic mention Google says it is designed to discount.

4. Structured and reference sources

Industry association member lists, chamber of commerce directories, licensing registers, conference speaker pages and government supplier lists are dull, static pages that rank quietly for years. They also feed the entity understanding that lets a system know your company is a distinct thing with a consistent description.

Entity consistency is the unglamorous foundation here. Same legal name, same description, same city, same URL, everywhere. If your company is described four different ways across four directories, every one of those pages is a slightly weaker signal. Our guide to getting a Google Knowledge Panel walks through the same entity work from the Google side, and the two efforts are the same effort.

A 90 day plan to get cited by AI search

Run it in this order. Each phase assumes roughly four to six hours a week from one person who knows the business.

Days 1 to 14: audit and unblock

1. Ask four assistants the ten questions your buyers actually ask, including your own brand name. Record every domain cited. That list is your real competitive set.
2. Open your robots.txt. Confirm OAI-SearchBot is not blocked unless you intend it to be, and confirm Googlebot is unrestricted.
3. Confirm your key pages are indexed and snippet-eligible. A nosnippet or max-snippet directive left behind from an old project will keep you out of AI features entirely.
4. Write down the exact company description you will use everywhere. One paragraph. Never varied again.

Days 15 to 45: claim and populate

1. Claim every review and directory profile on the cited-domain list from step one. Use the standard description.
2. Request reviews from your last ten satisfied clients, with a specific question rather than a generic ask. “What problem did we solve and what changed” produces quotable text.
3. Join the one or two communities where your buyers actually post. Read for a week. Then answer.
4. Publish or refresh the two pages on your own site that best answer your highest intent questions, structured answer-first.

Days 46 to 90: earn and measure

1. Pitch three podcasts and two trade publications in your vertical, each with a specific angle you can defend.
2. Answer source requests in your lanes, weekly, for the whole period.
3. Re-run the same ten assistant prompts from day one and compare the cited domains. This is your measurement. There is no reliable third-party rank tracker for AI citation, and Google itself cautions that no third-party tool has access to its ranking or AI systems.
4. Check the Search Console performance reporting for your AI surfaces and compare direction, not absolute numbers.

Expect the first movement somewhere between week eight and week sixteen. Review platforms move first because they are crawled often. Earned mentions take longest and last longest.

What this costs in Canada

All figures CAD. Platform pricing changes, so verify current rates directly with each provider before budgeting.

Run entirely in-house, the first 90 days costs almost nothing in cash and roughly 60 to 80 hours of owner or marketer time. That is the honest trade. If you would rather buy the time back, our website growth packages cover the on-site and entity half of this, and we will tell you plainly which parts you still have to do yourself, because the review requests and the podcast appearances cannot be outsourced to anybody who is not you.

For context on what the broader discipline costs in this market, our Canadian SEO pricing guide breaks down the ranges by scope.

Common mistakes

• Buying mentions. Press release syndication, paid “as seen in” badges and mention packages. Google says directly that seeking inauthentic mentions is not as helpful as it seems. You are buying a signal the system is built to discount.
• Adding llms.txt and calling it strategy. Google states it does not use llms.txt or AI-specific markup for Search. The file is harmless and free, so leave it if you have it, but it is not a lever.
• Blocking every AI bot. A single robots.txt line can remove you from ChatGPT search while doing nothing about training.
• Writing fake-sounding reviews into existence. Platform enforcement is better than it was, removal is retroactive, and the reputational cost of being caught is worse than having five honest reviews.
• Measuring with a tool that claims inside knowledge. Google explicitly warns that no third-party tool has access to its internal ranking or AI systems. Prompt tracking you run yourself is less impressive and more truthful.
• Treating one mention as the finish line. Citation is a frequency game. A single trade article does not move an answer that is assembled fresh each time.
• Spreading across eight communities. Shallow presence everywhere reads as spam and gets removed. One community, properly, is the move.

Worked example: a Calgary trades contractor

Hypothetical, to show the shape of the work rather than to claim a result. A 12-person electrical contractor asks why assistants recommend three competitors and never them.

The audit finds the cited pages are: a HomeStars category list, two Reddit threads in a Calgary subreddit, a city licensing page, and one competitor’s own service page. The contractor appears on none of them. Their website is fine. Their robots.txt blocks every bot with “AI” in the name, including OAI-SearchBot.

The fix, in order: unblock OAI-SearchBot, claim and complete the HomeStars and Google profiles with a consistent description, request reviews from twenty recent jobs asking what problem was solved, have the owner answer electrical questions in the local subreddit twice a week under his own name, and get listed on the provincial association member page. Nothing here is clever. All of it is off-site. The website work is already done, which is precisely why more website work was never going to change the answer.

Pros and cons of going off-site first

FAQ

Why are citations important in AI search optimization?

Because the citation is the only place your brand name appears in an AI answer. If the model answers the question and names three competitors, the buyer’s shortlist is formed before they ever reach a website. A citation is the modern equivalent of being on page one, except the page now holds three or four results instead of ten.

Do backlinks help you get cited by AI?

Indirectly. A backlink usually arrives attached to a mention, and the mention is the part that matters for citation. A link from a page nobody reads, bought in bulk, carries neither. We go deeper on the trade-off in link building versus topical authority.

Are most GEO and AEO experts legit?

Treat anyone selling a proprietary AI ranking method with heavy scepticism. Google’s own guidance tells publishers to be cautious of services claiming special AEO or GEO techniques and states that no third-party tool has access to its internal systems. The legitimate version of this work looks like ordinary good SEO plus genuine off-site presence, and an honest practitioner will tell you that.

How do I track AI citations for my brand?

Run the same set of 10 to 20 buyer questions across the assistants your market uses, on a fixed schedule, and log which domains get cited. It is manual and it is the most reliable method available. Third-party AI visibility tools can speed up the logging, but none of them can see inside the systems they measure.

Will brands still need website traffic if AI gives users the answer?

Yes, because the answer does not take payment. Assistants are increasingly good at the research phase and still useless at the transaction. What changes is the shape: fewer top-of-funnel visits, higher intent on the ones that arrive. Your site stops being a library and becomes a closing tool. We covered the measurement side of that shift in why pages rank but get no clicks.

Does schema markup get me into AI Overviews?

No. Google states that structured data is not required for generative AI search and that there is no special schema.org markup for it. Schema is still worth shipping for rich results and for machine-readable clarity, which is why we put it on every page we build, but it is not the entry ticket people sell it as.

How long before any of this shows up?

Review and directory changes can be reflected within weeks because those pages are crawled frequently. Earned mentions and community presence typically take two to four months to appear in answers, and longer to appear consistently. If someone promises AI citation in 30 days, they are selling something else.

Conclusion

The reason AI answers cite everyone but you is rarely a markup problem and almost never an llms.txt problem. It is that the pages those systems trust for your category are owned by review platforms, communities and publications, and you are not on them. Fix your technical eligibility first because it takes an afternoon. Then spend the next quarter earning presence on pages you do not control. That is slower, less satisfying, and the only thing that reliably works.

If you want the on-site and entity half built properly while you handle the reviews and the relationships, tell us what you are working with through our intake form and we will tell you which parts of this list you actually need and which you can skip.