A 2025 Pew Research Center study found that users clicked traditional results in 8% of Google visits with an AI summary, compared with 15% when no summary appeared. Links cited inside those summaries received clicks in only 1% of visits.
That changes what visibility means. A page can influence a decision without receiving the visit and developers increasingly control whether its information can be found, interpreted, and cited correctly.
AEO is not a replacement for SEO; it is an additional requirement on the publishing system.
Start with the retrieval path
Google says its generative search features still depend on core Search ranking systems and indexed web pages. It also says there are no special technical requirements, machine-readable files, or schema types required for inclusion.
The familiar foundation still matters: crawlable URLs, accessible text, internal links, canonical tags, accurate metadata, and structured data that matches the visible page. AI Mode and AI Overviews may issue several related searches before assembling a response, so a single prompt can produce multiple retrieval paths through a site.
For developers, the first AEO audit is straightforward: verify that important information can be reached and rendered without relying on blocked scripts, private APIs, or ambiguous redirects. Google documents the requirements for AI features in Search.
Model facts before adding markup
Schema can help search engines interpret a page, but it cannot repair inconsistent source content.
Product specifications, authors, publication dates, locations, prices, and availability should be governed fields in the CMS—not details copied independently into page content, metadata, JSON-LD, feeds, and APIs. A shared content model lets each delivery layer draw from the same source.
This architecture reduces a common failure mode: the page says one thing, the structured data says another, and an external profile contains an older value. AI systems can retrieve every version; they cannot resolve an inconsistency the platform created.
The practical pattern is simple: model a fact once, validate it, then publish it through the formats that need it.
Make crawler access an explicit policy
AI platforms use different user agents for different purposes. Search retrieval, user-requested access, and model training should not be treated as a single permission.
OpenAI, for example, distinguishes OAI-SearchBot, which supports discovery in ChatGPT search, from GPTBot, which publishers can block when excluding pages from potential model training. Perplexity separately documents PerplexityBot for search results and Perplexity-User for user-triggered access. (OpenAI publisher guidance, Perplexity crawler documentation)
Encode those decisions across robots.txt, CDN rules, firewalls, and page-level controls. Then check server logs. A policy file describes the intended behavior; request logs show whether the intended crawler can reach the intended content.
Measure representation, not only traffic
Rankings and referral sessions remain useful, but they do not show whether an AI-generated response represented the organization accurately.
Bing Webmaster Tools now reports citations, cited pages, grounding queries, and visibility trends across supported AI experiences. Google has also introduced dedicated generative AI performance reports for a subset of Search Console properties. (Bing AI Performance, Google Search Console announcement)
Pair those platform reports with a controlled prompt set. Track which pages are cited, whether important entities and specifications remain accurate, and where responses repeatedly retrieve outdated or incomplete information.
AEO becomes manageable when it is treated like the rest of the platform: structured inputs, explicit access rules, observable outputs, and regular verification.
Top comments (1)
{"content":"the 8% stat is wild tbh, feels like we're all just guessing at what 'good' content even means right now"}