We are a generative engine optimisation agency in Toronto

Generative engine optimisation is the work of structuring your content and your presence across the web so that AI tools cite your business when they answer a question. Where SEO gets you ranked in a list of links, GEO gets you cited inside the answer itself, which matters because a growing share of research now ends there rather than on a results page. The work has four parts. Content gets restructured so that each section answers one question completely and quotably, with real specifics rather than adjectives, because models extract passages rather than pages. Schema markup states your facts in a form the machine does not have to infer. Entity building makes your business name, category, location and services consistent everywhere you appear, since inconsistency is the most common reason a model declines to recommend a company it has read about. And authority work gets you mentioned on the sources these tools already trust, which for local businesses means associations and local publications as much as links. Across our client sites the pattern has held: the businesses being cited are overwhelmingly those that already had search authority and then made their content extractable. GEO is not a replacement for SEO. It is what you do with the authority SEO builds.

SEO optimises content to rank in a list of links. GEO optimises content to be cited inside an AI-generated answer. The two overlap heavily, since authority, links and technical health feed both, and in practice most of a GEO programme is work an excellent SEO programme would already be doing. The differences are specific. GEO rewards extractable writing: direct statements, real numbers, prices, timelines and constraints, in passages that stand on their own. Marketing language that reads acceptably on a web page gives a model nothing to quote. Entity consistency carries more weight than it does in ranking, because a model needs to be confident that the business on your site, the one in the directory and the one on the profile are the same business. Third-party description matters more, so what other sites say about you is part of the work rather than a by-product. And measurement is different: not positions and clicks, but how often you are cited for a set of questions, on which platforms, and whether what is said about you is accurate. Businesses building both at once compound their advantage, because the same page can rank in Google and be the passage an assistant quotes, and neither result costs anything extra once it is in place.

By building credibility signals a language model can verify, and by writing pages it can quote. In practice that means four things. Be mentioned consistently across the associations and publications the model already trusts, since a single mention proves little and a consistent pattern across sources reads as evidence. Write in direct statements that can be lifted without the surrounding paragraph, which is why price ranges, timelines, processes with steps and honest constraints get quoted while claims about quality do not. Implement full schema so what you do and where you operate is stated rather than implied. And keep the details identical everywhere: name, address, phone number, categories, services. ChatGPT works from a mix of training data and live search, so both the durable presence and the current web matter. Two practical notes. Check what it already says about you before starting, because an assistant repeating a service you dropped or an address you left is a bigger problem than being absent. And expect the first citations on specific, long questions, the ones a person asks in a sentence, rather than on the broad category question. Those specific questions are usually the ones with buying intent behind them anyway.

Consistency comes from three things working together, and a gap in any one of them produces the pattern most businesses see, where they are cited occasionally and then not at all. The first is content clarity. Pages need to answer specific questions in direct language that can be extracted without ambiguity, which usually means shorter sections, plainer sentences and actual figures. The second is entity consistency. Your business name, category, location, hours and service list need to read identically on your website, your Google Business Profile and the other sites that describe you, because a model presented with three versions of your business has no confident answer to give. The third is authority: being mentioned on sites these tools recognise as credible, which for a local business means industry associations and local press rather than link farms. A fourth thing is recency. Assistants that use live search favour pages that look current, so a page updated this quarter competes better than the same page untouched for two years. When the three are in place, citations become reliable rather than sporadic, and they hold across platforms because all three signals are platform-independent. We baseline a set of questions at the start so you can see that shift happen rather than take it on trust.

Perplexity searches the live web when it answers, which makes it the quickest of the assistants to respond and the best early indicator that GEO work is landing. It favours sources with real domain authority, clearly structured content and complete schema, and because it is fetching pages in the moment, technical health matters more than it does with tools leaning on training data. Pages that load fast, use a sensible heading structure, and put the answer near the top get surfaced and cited more often than pages that bury it. Three things we prioritise for Perplexity specifically. Topical depth in one area rather than thin coverage of many, since it tends to cite sources that clearly own a subject. Content that answers the exact question in the first two sentences of a section, because that is the passage it lifts. And consistent entity signals, so it can attribute the answer to a business it can identify. It also shows its sources, which makes progress measurable in a way that is unusual in this work: you can ask the questions your customers ask and see, week by week, whether you are in the answer, who is there instead, and what the citation says about you.

Yes, and ecommerce is where the shift is most visible, because people now ask assistants what to buy, which brand is better and where to buy it, often before they visit a single store. A brand cited in that answer reaches the buyer at the moment of decision, with the implied endorsement of the tool doing the recommending. The work for a store is specific. Product schema comes first, since price, availability, ratings and specifications stated in structured form are exactly what an assistant needs to name a product with confidence. Product copy has to carry real detail: materials, dimensions, sizing that maps to something a person understands, care instructions, what the item is actually for. Descriptions written to sound premium give a model nothing to work with. Reviews matter for volume, recency and detail, because assistants lean on them heavily when comparing options. Category-level content is the piece most stores skip, and it is what gets cited for the comparison questions that decide the sale. Consistency across your own site, your marketplace listings and any retailer that carries you is the last piece, since conflicting prices or specifications make a brand an unsafe answer. Brands starting now are competing for a channel that is still largely uncontested in most categories.

By tracking citations rather than rankings: how often your business is cited in AI answers for a defined set of questions, on which platforms, in what position within the answer, and whether what is said about you is correct. We baseline that at the start of every engagement by running a set of real customer questions across ChatGPT, Perplexity and Google AI Overviews, and recording who is currently named. That baseline usually produces two surprises. Most businesses are absent from questions they assumed they owned, and where they do appear, some of the detail is out of date. Then we track four things monthly: citation frequency, whether the citation is accurate, where in the answer it falls, and which competitors are being cited for the same questions. Competitor tracking matters because this is a relative game. Being cited in three of ten answers is good if the leader is at four and poor if the leader is at nine. The measurement has real limits and we are straight about them. Answers vary between sessions and users, so any single test is noise. What we report is a pattern across a fixed question set, tested the same way each month, which is the only version of this number that means anything.

In our client work, citations typically start appearing within 60 to 90 days for a business starting from a reasonable SEO baseline. The order is predictable. Tools that search live, Perplexity and Google AI Overviews, respond first, because they can pick up a new page as soon as it is indexed. Assistants leaning more heavily on training data move slower, since your presence has to accumulate across sources rather than appear on one. Specific questions come before broad ones: the sentence-long question about a treatment, a price or a comparison gets answered from a page that addresses it directly, while the bare category question is contested by national sites and takes far longer. Three things decide where you land in that range: your current authority, how competitive your category is in AI answers, and how quickly the entity and content work gets done. Trades, local professional services and specialist clinics often see movement sooner, because few competitors are doing anything at all. Finance, insurance and health take longer, because the models are more conservative about which sources they will name. We report from month one against the baseline question set, so you can see the pattern building rather than wait for a result, and the compounding usually shows after the first 90 days.

It brought in four real businesses that reached out, not just visits. Our own CRM records four inbound leads on attribute.media attributed to chatgpt.com since 20 August 2026: Anytime Fitness on 20 August, landing on our gym marketing page; Family TLC on 28 August, landing on our mental health marketing page; Troy City Fitness on 9 September, landing on the gym marketing page again; and Brain & Barrel on 22 September, going straight to the contact page. That is a count of businesses that got in touch, not a conversion rate, a close rate, or a revenue figure. The attribution comes from our CRM rather than Google Analytics: attribute.media does not yet fire a lead event inside GA4 when the contact form is submitted, so GA4 alone cannot confirm these as conversions. GA4 does show the traffic quality behind them. Over the ninety days to 22 September 2026, organic sessions from chatgpt.com averaged 3.1 pages per visit and 4 minutes 52 seconds, with 63% counted as engaged. This is one dated result, on one site, over one window. It says nothing about what another business or another vertical should expect from the same mechanism, only that it produced real enquiries here. Further results from client work sit on our clients page.

Yes, and our own mental health marketing page is the clearest proof of it. In Google Search Console that page sits at an average position of 35.9, with 947 impressions and zero clicks, over the ninety days to 22 September 2026. Over the same window it is one of the two pages on attribute.media that ChatGPT cites most, alongside our gym marketing page. Ranking and citation are measured by different things. A ranking position comes from Google weighing hundreds of signals against every page competing for a query. A citation depends only on whether a page answers a specific question completely enough for a model to lift the answer directly. A page can do that while still losing the broader keyword fight in Google, because the two systems are not judging it on the same question. What produces the citation is covered in our answer on what generative engine optimisation actually involves, above: content built to answer one question completely, real specifics rather than adjectives, and schema that states facts rather than leaving a model to infer them. None of that requires a high Google ranking first. That is why a page can be an early GEO result while the SEO work behind it is still catching up.

Five pages have received an organic AI citation on attribute.media over the ninety days we have been able to track it: the homepage, this page, the web development page, and the two vertical hubs, gym marketing and mental health marketing. Nothing else has, including the shorter guide articles elsewhere on the site that cover narrower topics in less depth. It is not that hub pages get cited and everything else does not. It is that these five pages are the ones on the site built as a single, deep answer to a topic, with direct FAQ content rather than a shorter overview that sends the reader elsewhere for detail. The guide articles are shorter by design, built to answer one narrower question well rather than to be the definitive page on their whole topic. A model has less reason to lift an answer from a page that only partly covers a topic when a fuller answer sits one click away on the same site. It is also why new GEO questions get added to the FAQ block on this page rather than spun off into a separate page: depth in one place gets cited more than the same content split across several thinner ones.

Six matter for a business today: ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot and Google AI Overviews, which sit at the top of ordinary Google results. They do not work the same way, and the differences change what to do first. ChatGPT, Gemini, Perplexity and AI Overviews run a live web search when they answer a recommendation question, so a new page can be cited within days of being indexed. ChatGPT searches on Bing's index plus its own crawler, which means a site blocked in Bing is invisible to it. AI Overviews draw on Google's own ranking, so a page that already ranks in the top results is the one most likely to be quoted. Perplexity shows every source it used, which makes it the easiest to measure. Claude and Copilot lean more on what the model already knows, so presence across the wider web counts for more with them than any one page does. Two things are true across all six. They quote the passage that answers the question most directly, and they trust a business they find described the same way in several independent places. Semrush's 100 million citation study found only about one site in ten is cited by both ChatGPT and Perplexity for the same questions, so a business cannot assume that being in one answer means being in the others. That is why we baseline and report each assistant separately.

Not on the evidence so far. An llms.txt file is a plain text index of a site written for AI crawlers, and it has been widely recommended. Ahrefs looked at 137,000 domains in May 2026 and found that 97 percent of the llms.txt files on them received no requests from any AI crawler at all. The AI retrieval bots accounted for about one percent of the requests those files did get. Google has said it does not support the file and is not planning to. The assistants read the ordinary HTML of a page, the same text a person sees. We keep an llms.txt on this site because it costs nothing and does no harm, but we do not present it to clients as a reason they get cited, because it is not one. The same goes for content that exists only inside structured data: in tests, facts that appeared only in schema and not in the visible text were almost never extracted. What does get read is the page itself. So the work that matters is the visible content: the answer under each question, specific detail a model can quote, and a site organised so that the information a client asks for is easy to find and easy to lift. That is a bigger job than adding a file, and it is the one that shows up in the answers.

It means designing the site around the questions clients ask, instead of around the pages a business wants to show. A conventional service site has a page per service, a page about the company and a contact page, each written to be read top to bottom. An assistant does not read it that way. It arrives with a specific question, scans for the passage that answers it, and leaves with that passage or nothing. So a repository site is built as a set of complete answers: what each service costs, how long it takes, who it is right for, what happens at the first visit, what goes wrong and how it is handled, which neighbourhoods are covered. Each answer sits under its own heading, in short sentences, with the specifics first. The pages are longer, not shorter, because depth in one place is what the assistants cite. On this site, the five pages that get cited are the five built this way. The shorter articles get nothing. In practice the rebuild usually means fewer pages than the client expects, each one far deeper, a question and answer block on every service page, and the business's facts, such as name, address, hours and services, stated identically on every page and every profile. It is the same information the business already gives out on the phone every day. The difference is that it is written down, organised and reachable.

They have started. ChatGPT added checkout inside the conversation for some United States merchants in late 2025, so a person can ask what to buy and complete the purchase without visiting the store. Google's AI Mode can book a restaurant table or a local appointment by calling the business on the user's behalf. Perplexity has its own shopping flow. These are early and uneven, and most of them are not yet available in Canada, but the direction is clear: the assistant is becoming the place where the decision and the action both happen. That raises the stakes for being the business it recommends, because the person may never see a results page, a website or a competitor at all. It also changes what a site has to contain. An assistant acting for a customer needs facts it can rely on: current prices, real availability, the exact services offered, the booking route, the address and hours. A business whose details differ between its site, its profile and the other sites that list it is a risk the assistant will route around. The work we describe on this page, being written about, pages built as answers, a consistent identity, is the same work that prepares a business for that. The more people understand what these tools are about to do for them, the more it matters who the tools recommend.