AI search monitoring
AI search monitoring tracks how answer engines like ChatGPT, Perplexity, Gemini and Google AI Overviews answer prompts about your brand and category: which brands they mention, in what order, and which pages they cite. Zenrows runs the chat interfaces in Browser and fetches AI Overview pages with Fetch, so you store and score every answer over time.
Updated
{
"prompt": "best project management software for agencies",
"engine": "perplexity",
"country": "us",
"run": 2,
"captured_at": "2026-10-16T09:41:07Z",
"answer_text": "For agencies, the tools recommended most often are ...",
"brands_mentioned": [
{ "brand": "Example Rival", "position": 1, "recommended": true },
{ "brand": "Your Brand", "position": 3, "recommended": true },
{ "brand": "Other Tool", "position": 4, "recommended": false }
],
"citations": [
{ "position": 1, "url": "https://www.example-reviews.com/best-pm-tools", "domain": "example-reviews.com" },
{ "position": 2, "url": "https://www.yourbrand.example/agencies", "domain": "yourbrand.example" }
],
"your_brand": { "mentioned": true, "position": 3, "cited": true },
"sentiment": "positive"
}
What AI visibility tracking measures
A fixed prompt set, run through each engine; the answers serve several jobs.
| Job | Data you collect | Typical sources |
|---|---|---|
| Brand visibility | Presence, position, share of answers | |
| Citation tracking | Cited URLs and domains, in order | |
| AI Overview tracking | Overview presence, text and sources per keyword | |
| Competitor share of voice | Mentions per competitor | The same prompts on every engine |
| Answer accuracy checks | Claims on your pricing, features, availability | Brand prompts like "Does Your Brand support SSO?" |
| Source gap analysis | Sites cited for your category that never mention you | Citations joined with content and PR targets |
Why AI answers are hard to track
A trendable number means driving a chat interface, passing its bot protection and sampling answers that change per run.
The answer streams in after submit; no URL to fetch
Hosted Playwright or Puppeteer in Browser types, waits for the stream, reads answer and sources
Cloudflare challenges and Turnstile
Zenrows' stealth in Browser gets through Cloudflare, Turnstile and similar
Missing from raw HTML, or different per market
Fetch with Adaptive Stealth Mode and geotargeting returns the page a local user sees
Different brands an hour later
Run each prompt several times; trend the share mentioning you
Mentions in paragraphs, citations in footnotes
Markdown keeps text and links parser-ready
How to monitor AI search with Zenrows
- Write the prompt set: category ("best CRM for startups"), comparison ("Your Brand vs Example Rival") and brand prompts, each with a country.
- Run the chat engines in Browser, saving each full answer and its source links.
- Fetch AI Overviews for your keywords per country with Adaptive Stealth Mode.
- Parse mentions and citations: brands and aliases in order, cited URLs as domains.
- Repeat and trend several times a week: share of answers, average position, citation share per engine.
# pip install playwright
from playwright.sync_api import sync_playwright
PROMPT = "best project management software for agencies"
with sync_playwright() as p:
browser = p.chromium.connect_over_cdp("wss://browser.zenrows.com?apikey=YOUR_ZENROWS_API_KEY")
page = browser.new_page()
page.goto("https://www.perplexity.ai/")
page.fill(PROMPT_INPUT_SELECTOR, PROMPT) # the chat input box
page.keyboard.press("Enter")
page.wait_for_selector(ANSWER_DONE_SELECTOR, timeout=90_000) # shown when the answer finishes
answer = page.inner_text(ANSWER_SELECTOR)
sources = page.eval_on_selector_all(SOURCE_LINK_SELECTOR, "els => els.map(e => e.href)")
print(answer, sources)
browser.close()# The weekly keyword set for one country, as a single job
curl -X POST "https://async.api.zenrows.com/v1/jobs" \
-H "X-API-Key: YOUR_ZENROWS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"type": "regular",
"status": "closed",
"zenrows_params": { "mode": "auto", "proxy_country": "us", "response_type": "markdown" },
"tasks": [
{ "url": "https://www.google.com/search?q=best+crm+for+startups&gl=us&hl=en", "external_id": "crm-startups|us" },
{ "url": "https://www.google.com/search?q=crm+with+email+automation&gl=us&hl=en", "external_id": "crm-email|us" }
]
}'npx -y @zenrows/cli init
zenrows fetch "https://www.google.com/search?q=best+crm+for+startups&gl=us&hl=en" --proxy-country us --output markdown
# keywords-us.jsonl: one {"url": "..."} per Google results page
zenrows batch create keywords-us.jsonl --proxy-country us --output markdown --wait# Connect once
claude mcp add zenrows -e ZENROWS_API_KEY=YOUR_ZENROWS_API_KEY -- npx -y @zenrows/mcp
# Then ask:
Use the Zenrows browser to open perplexity.ai, ask "best project management software for agencies", wait for the answer to finish, and list every brand it mentions in order and every source URL it cites. Close the session when done.Which Zenrows product fits
- Capture answers from ChatGPT, Perplexity or Gemini BrowserHosted Playwright or Puppeteer, protection handled
- Track AI Overviews for a keyword list FetchThe rendered Google page from your chosen country
- Run thousands of keyword-country pairs weekly BatchOne job per country, results by webhook
- Check by hand how engines describe you MCP ServerClaude or Cursor fetches and summarises the live page
What success looks like
| Measure | What good looks like | How to check it |
|---|---|---|
| Prompt coverage | Every prompt, engine and country, each period | Stored answers against plan |
| Sample depth | Weekly changes exceed run-to-run noise | Run one prompt twice in a day and compare |
| Capture completeness | Full text and cited URLs per answer | Flag answers missing citations |
| Mention accuracy | Aliases and product names caught, no false matches | Hand-review 50 answers monthly |
| Trend history | Twelve or more weeks of share per brand and engine | Check for gaps before each report |
AI search monitoring FAQ
-
Also called AI visibility or LLM tracking: asking answer engines your buyers' questions and recording how often and where your brand appears, whether your pages are cited, and how competitors compare. Rank tracking for answer engines.
-
Rank tracking records a fairly stable position among ten links. An AI answer may name five brands or none, cite pages that do not rank, and change between identical prompts, so you track share of answers and citations across runs.
-
Yes, through Browser: your Playwright or Puppeteer script submits the prompt in a Zenrows browser and reads the finished answer and sources. Review each engine's terms; keep volume proportionate.
-
Google publishes none. Fetch each keyword's results page from your country, with Adaptive Stealth Mode rendering when needed, and read the overview block and its links. Batch runs a keyword list per country as one job.
-
One weekly run is a weak signal. Run each prompt at least three times per period per engine and report the share mentioning you; daily during a launch or after publishing content meant to be cited.
-
Real buyer questions: category ("best tool for X"), competitor comparisons, problems leading to your category, brand facts on pricing and features. Start with fifty to two hundred per market, grouped by topic.
-
Run category prompts through ChatGPT search in Browser and count citations per domain across runs. The top domains (review sites, comparisons, documentation, forums) are what engines trust; those that never mention you are your outreach and content targets.