AI answers have changed where customers discover and compare brands. A buyer can now get a shortlist from an assistant before visiting a single website.
But some things haven’t changed. Customers still need trustworthy information, search systems still need to find it, and marketers still need to know whether visibility brings in qualified business. What’s harder now is measuring what happens before someone reaches your site.
One limitation up front: Google documents how its AI features work with Search, and OpenAI documents its crawlers, but no AI platform publishes a full account of how it chooses sources. The priorities below rely on what the platforms do document, and on work that helps buyers whatever the algorithm does.
Google says its AI features are built on its core Search ranking and quality systems, and that useful, original content and sound technical foundations still matter. To appear as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to show with a snippet. Google also says no special AI files, markup or content chunking are needed.
So keep working on crawlability, indexing, page experience, helpful content and genuine authority. Then widen your reporting beyond rankings and visits:
That’s a broader view of search performance, not the end of SEO.
Picture a software company with dozens of productivity blog posts but no clear page on its integrations, setup time, security controls or product limits. A buyer asks an assistant which tools work with their accounting system. The company has plenty of content and none of the evidence that buyer needs.
Another generic article won’t fix that. A clear integration page, a plain product explanation or an honest comparison will.
Go through your product, service, industry and comparison pages and ask what a serious buyer would still need to know. Add facts, practical examples, documented experience and limits. Don’t manufacture testimonials or case studies.
Google is explicit on one point. Creating separate pages for every variation of a question, including the related searches its AI features run, mainly to manipulate rankings or AI answers, breaks its scaled content abuse policy.
Fewer, better pages are both safer and more useful.
A great page does little for a system that can’t reach it. Have someone check robots.txt, index and snippet controls, HTTP responses, internal links, and any CDN or bot-protection rules. Test what legitimate crawlers can reach, not just how the homepage looks in your browser.
Be precise about what each crawler does. OpenAI’s three are documented in its crawler overview:
| Crawler or control | Used for | What restricting it means |
|---|---|---|
| OAI-SearchBot | ChatGPT search | Your pages can drop out of ChatGPT search answers, though they may still appear as navigational links |
| GPTBot | Possible OpenAI model training | A training preference only; it’s independent of OAI-SearchBot |
| ChatGPT-User | Some page visits a user triggers | Robots.txt may not apply, and it doesn’t decide search inclusion |
| PerplexityBot | Surfacing and linking sites in Perplexity search | Your pages can drop out of Perplexity results |
| Perplexity-User | Pages fetched to answer a user’s question | Generally ignores robots.txt, because a user asked for the fetch |
| Googlebot | Google Search indexing, including AI Overviews and AI Mode | Can reduce eligibility for Google’s Search experiences |
| Google-Extended | Some Gemini training and grounding uses | Doesn’t affect inclusion or ranking in Google Search |
If you use Cloudflare, check your actual zone settings before blaming your content. Its AI crawler controls now separate search, agent and training crawlers, and some combinations of settings can block more than teams expect, including multi-purpose crawlers like Googlebot and Bingbot on certain pages.
Your website isn’t the only source customers or AI systems read. Reviews, industry publications, partner directories, comparison sites and public documentation all shape what’s known about you.
The goal isn’t planting mentions. Google says seeking inauthentic mentions isn’t as helpful as it might seem, because its spam systems still apply. The goal is making correct, useful information available where customers already do their research.
Start with consistency. If your name, features, pricing model and service regions differ across major sources, buyers get confused long before any assistant does.
When partners or publications cover you, give them substance: accurate documentation, the method behind any research you share, and examples customers have approved.
First-party reporting got more useful in 2026, and it’s easy to blend unrelated numbers into a misleading dashboard. Keep them apart:
| Source | What it can tell you | Main limitation |
|---|---|---|
| Search Console, generative AI performance report | How your pages appear in Google’s AI features, by page, country, device and date | Doesn’t list the questions people asked |
| Bing Webmaster Tools, AI Performance | Citations, cited pages and grounding queries in Microsoft’s AI experiences | Citations aren’t visits or rankings |
| GA4 referral and channel reports | Visits from identifiable AI assistants | Some app traffic loses attribution; Google AI feature clicks appear as Google organic |
| AI prompt monitoring | Mentions, citations, competitors and accuracy for chosen questions | A chosen sample, not market-wide share |
| CRM and conversion data | Leads, sign-ups, opportunities and revenue | An AI answer’s influence often isn’t visible |
One practical GA4 setup. Create a custom channel group for AI assistant referrals, with a source rule such as:
^(chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai)$
Place it above Referral, because GA4 assigns traffic to the first channel it matches in the group’s order. Then check which sources actually appear in your property and adjust the rule. Standard GA4 properties allow only two custom channel groups, so plan before you use one.
Above all, an AI mention, a citation, a click and a qualified lead are four different events. Keep them separate before deciding whether things are improving.
Keyword research still matters, but it doesn’t capture how a customer explains a decision to an assistant. The keyword might be “accounting software”. The real question is “Which accounting software handles inventory and GST for a small wholesaler in India?”
That question adds a business type, a location and specific requirements. It also tells your content team exactly what’s missing.
Build a manageable list from sales calls, support tickets, site search, search data and buyer research. Group it by problem, requirement, comparison and branded intent. Then compare answers across the platforms your customers actually use, and look for patterns that repeat, not single screenshots.
The valuable step comes after: giving the right page, fact or correction to someone who can fix it.
You don’t need a GEO department to start. You need clear owners across SEO, content, marketing and development, and a shared way to decide what’s next.
We’d stop publishing thin pages for every phrasing of a question. We’d stop treating an optional file like llms.txt as a lever for Google. We’d stop calling one AI response a “ranking”. And we’d stop celebrating a visibility percentage without knowing which questions it measured, or why those questions matter to buyers.
The strongest AI search strategy for 2026 is a practical one: keep SEO healthy, make sure crawlers can reach you, publish the facts that help people choose, measure honestly, and turn real gaps into work someone owns.
Compare your AI visibility with your search and customer data, then pick three improvements that would also make your site better for a human buyer.
CHECK YOUR AI VISIBILITYSee where you stand, uncover visibility gaps, and discover what to improve next.
Get your first visibility score in minutes.