Loading
AI assistants recommend businesses by reading pages that answer a question directly and by cross-checking claims against third-party sources. You influence that with extractable answers, structured data, and citations you do not control. Here is what actually works.
Updated · 2 min read
Classic search returns a list and lets the user choose. An assistant returns an answer and names one or two businesses, so second place is invisible rather than merely lower.
The inputs overlap with SEO, because assistants lean on search indexes and on crawlable pages, but the winning format differs. Search rewards a page that covers a topic. An assistant rewards a passage it can lift and attribute without hedging.
The single highest-return change is structural: put the answer in the first two or three sentences of the page, with the numbers in it, and let the detail follow.
That is why every guide on this site opens with a short summary paragraph rather than a preamble. If a model has to infer your answer from six paragraphs of context, it will usually quote a competitor who simply stated it.
Structured data does not make you rank, but it removes ambiguity about what you are and where you operate. The ones that matter: an organization or local business entity with your address and service areas, FAQ markup on real question and answer pairs, breadcrumbs for hierarchy, and article markup on guides.
A plain text summary at /llms.txt is also worth publishing. It costs an afternoon, states your facts without the marketing wrapper, and gives a crawler something unambiguous to read.
This is the uncomfortable half. Assistants corroborate what a business says about itself against sources it does not own: directories, review platforms, industry listicles, and press. A site can be perfectly optimized and still go unrecommended because nothing else on the web confirms it exists.
So the off-site work is not optional. Claim and complete the directory profiles that carry weight in your industry, collect real reviews on the platforms buyers actually cite, and get named in roundups where you genuinely belong. Realistically, on-site work is a minority of the outcome.
Ask the assistants directly, on a schedule, in the phrasing a customer would use, and record what comes back. It is crude and it is the only feedback loop currently available, because there is no analytics panel for AI answers.
Then watch the durable proxies: branded search volume, direct traffic, and referrals from the sources assistants tend to cite. If those move, visibility is moving with them.
No. It is a proposed convention, not a standard that assistants are obliged to honour. It is cheap to publish and unambiguous when it is read, which is the whole argument for it.
Only if you do not want to be recommended. Publishers protecting paid content have a real case for blocking. A business that wants to be found does not.
Assume months rather than weeks. Model training and retrieval indexes update on their own schedule, and the off-site corroboration is the slow part.
Put the first dollar into Google Ads if you need customers this quarter and your site already converts; put it into SEO if your margins cannot carry a cost per click and you can wait 4 to 6 months for results. Most small businesses should run a small search campaign first, because it shows in weeks which searches are worth years of SEO.
EcommerceGoogle's Core Web Vitals thresholds are a Largest Contentful Paint of 2.5 seconds or less, an Interaction to Next Paint of 200 milliseconds or less, and a Cumulative Layout Shift of 0.1 or less, measured at the 75th percentile of real visits. On most Shopify and custom stores the failures come from images, third-party scripts and fonts, in that order, and fixing those three is a few weeks of work rather than a replatform.
RegulationLaw 25 applies to any business that collects personal information from people in Quebec, whether or not it has an office there. Since September 22, 2023 that means a published privacy policy, a named privacy officer, consent that is specific and separate, and tracking that stays off until the visitor turns it on. Administrative penalties reach the greater of $10 million or 2 percent of worldwide turnover.
Nockta builds ecommerce, custom software, and Amazon operations for businesses across Ontario and Quebec. See where we work or the full service list.