AEO by the Numbers: What Research Shows About AI Answer Engines

Answer engine optimization only makes sense if the numbers behind it are real. This brief pulls together figures you can check yourself: how many adults use AI chatbots, how unstable ChatGPT's local recommendations can be, what a major research paper found when it changed page content, and what Google publishes about AI Overviews. Every figure below links to its source. If we could not verify a claim, we left it out.

If you want the plain definition of AEO first, read What is AEO. For the name debate, see AEO vs GEO. To try a free, very basic check of your own URL, use the basic AEO audit tool.

Key numbers at a glance

Treat every row as a measurement from a specific study or survey — not a promise that your market will behave the same way.

1. More people ask chatbots for information

Pew Research Center's Americans and AI 2026 report (survey of 5,119 U.S. adults, Feb 17–23, 2026) found that 49% of U.S. adults say they use AI chatbots like ChatGPT, Gemini or Copilot — up from 33% in 2024. Roughly one in four use these tools daily, according to the same report's summary.

Searching for information is the most common use Pew measured: 42% of U.S. adults say they use chatbots to search for information. Among employed adults, 38% say they use chatbots for tasks at work (Pew chart).

An earlier Pew short read (June 2025) reported that 34% of U.S. adults had ever used ChatGPT, roughly double the share from 2023 (Pew). Separately, in August 2025 Pew found that 65% of U.S. adults at least sometimes come across AI summaries in search results (Pew).

None of these surveys prove that your next customer will ask ChatGPT before calling you. They do show that chatbot and AI-summary exposure is no longer a niche behavior in the United States.

2. Local AI recommendations churn quickly

For local service businesses, the practical question is not "does AI exist?" It is "if ChatGPT names me today, will it name me next week?"

Spearleaf's August 2026 study asked ChatGPT the same "who is the best {service} in {city}" question across 216 U.S. markets (12 services × 18 cities), then asked again 4–9 days later. In 214 paired markets with valid panels both times:

• 48.2% of businesses shown in wave one were still shown in wave two (469 of 974).
• The same #1 held in only 34.1% of markets (73 of 214).
• Mean panel overlap was 0.36 Jaccard (1.0 = identical lists).
• Google's traditional local pack in the same window was much steadier: 0.82 Jaccard (n=198).

Primary write-ups: study page and half-life insights. The study describes one pinned model over one week in August 2026 — a measurement of that window, not a law of nature.

The takeaway for AEO measurement is blunt: one screenshot is one roll of the dice. An AEO visibility audit re-asks the same customer questions on a schedule for exactly this reason.

3. What the GEO research paper actually tested

The term "generative engine optimization" was popularized by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande in GEO: Generative Engine Optimization (arXiv:2311.09735), later accepted at KDD 2024. The paper introduces GEO-bench (about 10,000 queries across domains) and tests content edits against generative-engine-style answers.

Their headline result: proposed GEO methods can boost visibility by up to 40% on diverse queries in their evaluation. Top-performing methods in their write-up — cite sources, quotation addition, and statistics addition — achieved roughly 30–40% relative improvement on a position-adjusted word-count metric and 15–30% on a subjective impression metric (paper HTML). On Perplexity.ai as a real-world generative engine, they report visibility improvements up to 37%.

Important caveats the paper itself implies and that practitioners often skip:

• The ~40% figure is an upper bound for particular methods and domains, not an average you should put on a sales page.
• Effects varied by domain; the authors argue for domain-specific optimization.
• Some traditional SEO-style edits did not help and could reduce visibility in their tests.

In day-to-day AEO work, that maps cleanly to answer-first pages with specific facts, cited sources and honest FAQs — not keyword stuffing. See how to write answer-first service pages.

4. What Google says about AI Overviews and AI Mode

Google's Search Central documentation on AI features and your website is unusually plain:

• Best practices for SEO remain relevant for AI Overviews and AI Mode.
• There are no additional requirements to appear in those features, and no special AI-only optimizations required.
• To be eligible as a supporting link, a page must be indexed and eligible to show in Google Search with a snippet.
• You do not need special schema, machine-readable AI files, or new markup solely for these features.

Google also notes that meeting requirements does not guarantee crawl, index or serve. Site owners can manage inclusion in Search generative AI features via Search Console's Search generative AI control (rolled out widely as of August 31, 2026 per Search Console Help).

For AEO practitioners: fix crawl/index/snippet eligibility first. Fancy "AI SEO" add-ons are optional; eligibility is not.

5. Crawlers you actually need to allow (and which are separate)

Answer engines that fetch the live web need a path through robots.txt and your CDN. Documented agents include:

• OpenAI: OAI-SearchBot (search), GPTBot (training), ChatGPT-User (user-triggered fetches; robots.txt may not apply the same way).
• Perplexity: PerplexityBot (search indexing), Perplexity-User (user-triggered).
• Anthropic: ClaudeBot (training), Claude-SearchBot (search), Claude-User (user-triggered).
• Google: Googlebot and related crawlers; Google-Extended is a separate control for some Gemini training/grounding uses.

Training bots and search bots are often separate. Blocking training does not always mean blocking search — and the reverse. Details and examples: AI crawlers and robots.txt.

6. What this means for a local or mid-sized business

Put the sources together and a practical checklist emerges:

1. Assume some buyers will ask an assistant before they call — Pew's search-for-information share is large enough to matter.
2. Measure with repeats, not screenshots — Spearleaf's churn numbers make a single "we're #1" claim unreliable.
3. Write pages engines can quote — specific facts, citations, stats and clear answers matched the GEO paper's stronger methods.
4. Stay eligible for Google Search — indexed, snippet-eligible pages are the floor for AI Overviews.
5. Let the right crawlers through — blocked search bots never retrieve you.

That is the core of what AEOswitch does. The free basic AEO audit checks a few on-page basics. The paid $49 full visibility audit runs multi-engine repeated questions, builds a citation log and a ranked fix list — see plans or contact for the audit.

7. Mentions outside your website still matter

When answer engines retrieve sources for "best X in Y" questions, the cited URLs are often not the business homepage. They are directories, review platforms, local news, comparison posts and community threads. Spearleaf's study discussion of retrieval overlap (retrieved-domain Jaccard around 0.20 between waves in their paired sample) is a reminder that the source set moves — not only the shortlist of business names.

Practical implication: cleaning your own site is necessary and not sufficient. Look at which domains appear in citations for your category, then earn honest presence there. That is slow work. It is also why Ongoing AEO retainers exist.

8. A measurement cadence that matches the data

Given Pew-scale adoption and Spearleaf-scale churn, a workable cadence for most local and mid-sized businesses looks like this:

AEOswitch's $49 Audit is the baseline row. Audit + Fixes and Ongoing cover the rest. The free tool only covers a thin slice of on-page checks.

Limits of this brief

We did not invent any of the percentages above. We also did not re-run the studies. Industry blogs often recycle "40% visibility" without the upper-bound context; we keep that context. Vendor case studies without methods were excluded. If a figure cannot be opened and checked on a public page, it does not belong here.

We will update this brief when major public studies or platform docs change. If you spot a broken source link, email get@justinzwick.com.

Want the human version of this baseline for your business? Book the $49 audit, browse free tools, or call (719) 937-3354.