The Death of SEO, The Birth of AEO (Answer Engine Optimization)
How AI answers change discovery, what publishers can do, and why citations still need to be measured alongside traffic and conversions.
When I wrote about this in January 2025, I expected AI answers to change the role of aggregators and search traffic. By April 2026, people could use AI services to get answers and complete more tasks without first visiting a list of websites.
Google says its existing SEO requirements apply to AI Overviews and AI Mode, with no special schema required1. For publishers, the practical question is whether these services can find accurate information on your site and whether the resulting visibility helps your business.
1. Aggregators still have a role
In January 2025 I wrote that aggregators had infrastructure for autonomous booking but lacked trust, regulations, and user comfort. That gap has not simply disappeared. Expedia's April 2026 research found that travelers still preferred trusted travel brands when booking, even as AI offered help with planning2.
KAYAK introduced KAYAK.ai in 2025 as a test lab for conversational travel search using its travel data3. That is a change in how people search an aggregator, not evidence that aggregators have become unnecessary or that their economics are unchanged.
For aggregators, one response is to make their data useful and accessible to AI services, rather than relying entirely on direct visits.
2. What AI visibility requires
AEO focuses on whether an AI service uses and cites your information. SEO also considers visibility and clicks in search. For Google’s AI features, the underlying SEO requirements still apply1.
Keep product facts current and consistent across the places that publish them. This gives readers reliable information, but does not guarantee that a model will cite your site.
Do not assume schema completeness replaces other search work. Google says existing SEO best practices remain relevant to its AI features1.
3. Practices to test on your site
I tested these practices with a few teams in 2025. They are useful starting points, but I do not have controlled results here that establish their effect on citations or conversions.
Lead with the answer
Lead with a direct answer and add context where readers need it. There is no fixed word count that guarantees a citation.
Avoid a long generic introduction before answering the question. Put the answer first, then explain the conditions and evidence behind it.
FAQ schema, done right
Organize related questions so readers can find the answer they need. Answer directly, with enough context to preserve the meaning. Check that any markup accurately describes the visible content.
Google does not require special schema for AI Overviews or AI Mode. If you add structured data, keep it consistent with the content users can see1.
Entity consistency across the web
Check the key facts across your About page, founder's LinkedIn, G2 listing, Crunchbase profile, and Wikipedia entry where one exists. Names and descriptions should agree. Inconsistency can confuse readers; there is no established rule that one mismatch makes a model stop citing you.
Keep product information usable
Write useful product descriptions for people and provide accurate structured data where it applies.
4. How this differs from app store optimization
In 2025 I argued that AEO was different from app store optimization. The distinction is still useful: app store optimization helps people find an app within a store, while AI search services can draw on information from many websites.
An AI agent may also help complete a transaction, depending on the service and tools it can access. Finding a service and successfully booking a flight, ordering a product, or scheduling an appointment are separate capabilities.
Anthropic reported a 72.5% success rate for Claude Sonnet 4.6 on OSWorld-Verified in February 20264. That is a computer-use benchmark result, not proof that it can reliably complete every booking or office task.
5. Write for conversational questions
In 2025 I wrote that conversational AI could support more complex requests than voice search. For content teams, this means considering how people describe their needs, rather than matching only a short keyword phrase.
For one client, I counted 47 ways someone might ask the same question. Those variations can help you check whether an answer covers the intended meaning. They do not each need a separate page or a repeated version of the same answer.
6. Metrics to track
Alongside rankings, traffic, and conversions, consider tracking:
- AI citation frequency. How often you're the source.
- Answer accuracy when cited. When AI quotes you, is it right?
- Structured-data accuracy. Does your markup match the visible content?
- Consistency of key facts. Do your listings and pages agree?
- Response latency, if you expose an API. Can it respond within the calling service’s time limit?
These measures help assess AI visibility and whether services can use your information. Compare them with visits and conversions to understand their practical value.
What to change
AI answers and agents create additional ways for people to discover and use a service. My recommendation is to make your information accurate and accessible, then measure whether AI referrals and citations lead to useful visits or completed tasks. Continue checking search traffic and conversions rather than assuming they have stopped mattering.