AEO Implementation: A Three-Stage Validation Method for Performance Evaluation

The biggest difference between AEO and traditional SEO is that you cannot directly see rankings after implementation.

SEO has Google Search Console and Ahrefs, which provide clear ranking data. However, AEO lacks these. ChatGPT won't tell you "how many times your website has been cited," and neither will Perplexity.

After you've implemented robots.txt, JSON-LD, and FAQ Schema and rewritten your articles into an AI-friendly structure, what's next? Are these efforts effective? Customers won't proactively tell you, "I heard about your brand through ChatGPT."

Therefore, we have independently developed a three-stage validation method: manual queries, bot traffic, and referral traffic. From the quickest to the most time-consuming, and from the lightest to the deepest approaches, validate your performance step by step.

Stage 1: Manual Queries to ChatGPT / Perplexity

This is the simplest method, taking about 10 minutes to complete.

Select 3 to 5 questions that "customers might ask before making a purchase," and then ask ChatGPT, Perplexity, Claude, and Gemini separately. Observe whether your brand or links appear, or if you are cited as a reference source.

The key is to ask questions from the customer's perspective. Don't ask from your own perspective, like "How to find Heiso," but rather think from the customer's point of view.

Some practical tips:

• Test individually across different models: ChatGPT and Perplexity reference different sets of websites. Perplexity directly lists reference sources, making it easier to track; however, ChatGPT won't reveal them unless you ask "What are the sources?"

• Rephrase questions in different ways: For example, using "digital transformation consultant," "AI implementation team," "B2B tech consultant," etc. Different phrasings often yield vastly different results.

• Keep detailed records: We recommend testing monthly and recording the results in a Google Sheet. Long-term observation is needed to see changes.

Limitations of this stage: It is subjective and difficult to quantify, and each answer may vary. However, when your website starts to be cited, this stage will be the quickest to show changes.

Stage 2: Bot Traffic Logs

For AI to cite your website, it must first crawl it. Logs will honestly tell you which bots are crawling and who they are.

Mainly focus on the following User-Agents:

• GPTBot — OpenAI's training crawler

• OAI-SearchBot — OpenAI's real-time search crawler (AEO directly impacts this)

• PerplexityBot — Perplexity

• ClaudeBot — Anthropic

• Google-Extended — Google Gemini's training crawler

• Bingbot — Shared by Bing and Copilot

Three tools are available for checking:

• nginx / Apache access log + grep User-Agent: The most原始 method with high flexibility, but requires writing commands yourself.

• Cloudflare Security → Bots: Free and quickly shows bot traffic percentage, but AI bot classification is relatively rough.

• Dedicated AEO Dashboard: Can classify AI bots, separate them from human traffic, and show which pages ChatGPT has crawled.

Heiso has independently developed the third type of dashboard and provides it as a standard feature for B2B official websites to clients. The /admin/visits interface is as follows:

Its features include:

• AI Crawler Classification: Automatically identifies and groups over 15 crawlers, such as GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, Bingbot, Applebot-Extended, etc.

• Real-time AI Assistant Fetch Detection: Identifies ChatGPT browse / Microsoft Copilot / Bing Chat requests from data center IPs and displays them separately from general visitor traffic.

• Pages Browsed by Each Bot: Directly verifies "what content AI is interested in on your site."

• Separate Display of Human and Bot Pages: Ensures human traffic statistics are not skewed by bot traffic.

This dashboard was developed by Heiso to address the pain point of "not knowing the effectiveness" when self-implementing AEO. Clients who "implement AEO but don't know the results" find this extremely useful. You will for the first time actually confirm ChatGPT's level of interest in your website, rather than guessing.

This stage observes "the number of times AI reads your website," not "the number of times AI cites your website." Being crawled doesn't guarantee being cited, but not being crawled means it's impossible to be cited. If there is traffic at this stage, it means the infrastructure has been validated.

Stage 3: AI Referral Traffic

This is the most rigorous, time-consuming, but also the most valuable metric.

When AI not only cites you but also provides a link, and users click to visit, these visitors will appear in your Analytics referral traffic report.

Google Analytics, Plausible, or self-built Analytics can all be used to view this; referral sources usually appear as follows:

• chatgpt.com

• perplexity.ai

• claude.ai

• gemini.google.com

If it was zero for several months, then starts to increase to once a week, then three times a week, then 10 times a month, it means your AEO is truly starting to work.

One thing to note: Although AI referral traffic is small in volume, its conversion rate is more than 5 times higher than Google Organic [Reference: /blog/aeo-vs-seo-traffic-comparison]. Don't dismiss it as insignificant just because there are only 20 visitors. These 20 people are all potential customers actively seeking solutions to their problems.

Important Concepts to Clarify in Advance

Even with all measures taken, it typically takes several months to see significant results.

AI model update cycles are measured in months, and the ranking mechanism for citation sources is not public. New content usually takes 4 to 8 weeks to be stably cited. Especially when your brand is still an "unfamiliar name" to AI, it may take even longer.

Don't give up after two weeks, saying "ChatGPT still didn't mention us." The feedback cycle for AEO is measured in weeks to months, not days.

Summary

In the SEO era, Search Console told us our rankings, but in the AEO era, there is no central dashboard. However, you can build your own. Manual testing is the dashboard, bot logs are the engine indicator lights, and referral traffic is the odometer.