How to check if Google AI Overviews recommend your product

To check whether a Google AI Overview recommends your product, ask the questions a buyer types while deciding — "best X for Y", "X alternatives", "X vs Y" — and look for your name in the overview. Then do the part that decides the outcome: read the sentence around the mention, and open the cited page beneath it. Being listed is not being recommended, and on a search-grounded surface the difference usually traces to one page.

The manual way (and why it stops working)

Search the way a buyer searches, not the way a marketer does. "Best help desk software for a small SaaS support team." "Acme alternatives." "Is Acme any good for a ten-person support team?" When Google shows an AI Overview, find your name, read the sentence it sits in — that sentence is the actual result — then expand the reference chips beside the block, because one of them is where the sentence came from. Six phrasings, fifteen minutes, no cost, and you'll know roughly where you stand today.

The trouble starts on the second pass.

• Buying intent fans out. "Best", "cheapest", "alternatives to", "X vs Y", and one variant per team size and use case — each is a separate search with its own overview, its own shortlist and its own framing of you. Covering that honestly is a morning, and it's a morning you owe again next month.
• Half your questions won't have an overview at all. Google doesn't answer every search inline, and a query with no overview tells you nothing about whether you're recommended — yet by hand it looks identical to an overview that simply left you out. Skip that distinction and every "not recommended" tally you report is quietly inflated.
• The wording moves between searches. An AI Overview is rebuilt from a fresh search each time, so Monday's flattering line and Friday's lukewarm one can both be real. One reading can't separate a genuine slip from ordinary churn.
• You will read yourself into a conclusion. Skimming an overview for your own name, "also worth considering" files itself under win with alarming ease. Extracting the sentence and judging it deliberately is a different exercise from scanning for a highlight.
• Nothing is stored and nothing is scheduled. In two months you'll want to know whether the framing improved after you rewrote the comparison page. A browser tab is not a dataset, and it can't page you the week a competitor takes your slot in the overview.

Do it with one API call

One POST to AgentGEO returns the raw record for a query: the full overview text and every source cited, structured. No headless browser, no chips to expand by hand — the managed-scraper engine is maintained for you, and of the six engines this is the densest citation surface. Its sibling, Google's AI Mode, is a separate surface with its own google_ai_mode key.

curl -X POST https://api.agentgeo.org/v1/fetches \
  -H "Authorization: Bearer ag_live_..." \
  -H "Content-Type: application/json" \
  -d '{"query": "Acme alternatives for help desk software", "surfaces": ["google_ai_overview"], "country": "US", "language": "en"}'

Now write the check that matters. Instead of testing whether the brand is merely present, split the overview into sentences and keep the ones your product appears in — then print the sources beside them, so the framing and its evidence stay attached. And keep the queries that returned no overview in a list of their own, because they are not evidence against you:

{
  "id": "run_9f2a7c41",
  "query": "best help desk software for a small SaaS support team",
  "surfaces": ["google_ai_overview"],
  "status": "completed",
  "recordsDelivered": 1,
  "creditsCharged": 1,
  "answers": [
    {
      "surfaceKey": "google_ai_overview",
      "status": "delivered",
      "answerText": "For a small SaaS support team, Acme is frequently recommended for its shared inbox and quick setup, while Globex is better suited to teams that need deep automation. Initech is a lower-cost option with a lighter feature set.",
      "sources": [
        { "title": "Acme Help Desk - Features", "url": "https://acme.example.com/help-desk", "position": 1 },
        { "title": "Best help desk software for small teams", "url": "https://example.com/best-help-desk", "position": 2 },
        { "title": "Globex vs Acme for support teams", "url": "https://globex.example.com/compare/acme", "position": 3 }
      ],
      "fetchedAt": "2026-07-18T09:14:22Z",
      "latencyMs": 61840
    }
  ]
}

That structure is what the script leans on. This one splits each overview into sentences, keeps the ones naming your brand, prints the sources behind them, and — the part specific to this surface — logs every query that produced no overview separately, so it can never contaminate the count:

import re
import requests

KEY = "ag_live_your_key_here"
BRAND = "Acme"

# Buying intent - the questions asked right before a decision.
QUERIES = [
    "best help desk software for a small SaaS support team",
    "Acme alternatives for help desk software",
    "Acme vs Globex for customer support",
    "cheapest help desk tool with a shared inbox",
]

MENTION = re.compile(rf"\b{re.escape(BRAND)}\b", re.I)


def sentences(text):
    parts = re.split(r"(?<=[.!?])\s+|\n+", text)
    return [p.strip(" -*\t") for p in parts if p.strip()]


overviews = 0   # queries that actually produced an AI Overview
named = 0       # of those, how many named the brand
missing = []    # queries with no overview - NOT "not recommended"

for query in QUERIES:
    resp = requests.post(
        "https://api.agentgeo.org/v1/fetches",
        json={
            "query": query,
            "surfaces": ["google_ai_overview"],
            "country": "US",   # search gl=
            "language": "en",  # search hl=
        },
        headers={"Authorization": f"Bearer {KEY}"},
        timeout=200,  # an AI Overview SERP round-trip runs 40-90s
    )
    resp.raise_for_status()

    for answer in resp.json()["answers"]:
        if answer.get("status") == "failed":
            # No overview on this SERP: zero credits, and not the same thing
            # as an overview that recommended someone else. Log, don't score.
            missing.append(query)
            continue

        overviews += 1
        text = answer.get("answerText") or ""
        hits = [s for s in sentences(text) if MENTION.search(s)]

        print(f"\n{query!r}")
        if hits:
            named += 1
            for line in hits:
                print(f"  says: {line}")
        else:
            print("  named someone, not you")

        # SERP-grounded: a delivered overview almost always cites. Open these.
        srcs = sorted(answer.get("sources") or [], key=lambda s: s["position"])
        for s in srcs:
            print(f"    {s['position']:>2}. {s['url']}")

# Two denominators, never one: coverage, then share within coverage.
print(f"\noverview shown for {overviews}/{len(QUERIES)} queries")
if overviews:
    print(f"named in {named}/{overviews} of the overviews that ran")
if missing:
    print(f"no overview (log, don't score): {', '.join(missing)}")

The output isn't a score, it's evidence: for each buying query that produced an overview, the exact sentences Google wrote about you and the pages it wrote them from. Four queries against one surface is at most four credits — a query with no overview comes back failed and costs nothing, which is why missing is a list rather than a silent skip. An ag_test_ key returns labelled demo records at zero credits while you shape the parser.

Read the sentence and its citations together and the fix usually names itself:

Let your agent do it

This job suits an agent better than a script, because the interesting step is the second one: fetch the overview, then go read the page that produced it. One command connects the server:

claude mcp add agentgeo -- npx -y agentgeo-mcp --key ag_live_...

That's the install: agentgeo-mcp over stdio, zero npm dependencies, Node.js 18+, and a single tool called fetch_raw_answers with query and surfaces required and country defaulting to US. From there you describe the task rather than implement it:

Fetch the google_ai_overview for "best help desk software for a small SaaS support team". If there's no overview, say so and stop. Otherwise quote the sentence that mentions Acme, then open the source it most likely came from and tell me whether that page actually says it.

Notice where the judgment sits. AgentGEO hands back the record — the overview text and its sources — and stops; your agent, which knows your positioning and what you'd call a fair description, decides what counts as a recommendation and whether the cited page needs correcting or out-writing. It also knows the difference between "no overview" and "not named", and won't score the first as the second. That split is the whole argument for a data layer rather than a dashboard; if you'd rather drive it from a script, the Python and curl paths call the same endpoint.

Common problems

Five things that will make this check lie to you, in rough order of how often they do:

• A missing overview is not a rejection. When Google shows no AI Overview, the record returns status: "failed" with "Google returned no AI Overview for this query" at zero credits. Log those queries in their own list. Fold them into a "not recommended" bucket and every recommendation rate you publish is understated.
• A failed overview is not a snapshot to redeem. The chatbot surfaces hand back a providerFields.snapshot_id on a slow scrape that you can retry against. AI Overviews come through a different SERP fetch path, so there's no snapshot id here — re-run the query later instead, and accept that it may honestly fail again.
• Substring matching over-reports. A bare "acme" in text also fires on "AcmeBank" and "the acme of support". Use a word-boundary regex and keep an ignore list for the namesakes already living in your category, or you'll count mentions that were never you.
• Branded queries flatter you. "Acme alternatives" contains your name, so an overview that appears will almost always mention it — the query guaranteed it. Keep branded and unbranded queries in separate buckets, or your recommendation rate mostly reflects your own query set.
• Locale changes the overview and its citations. country sets the search gl= and language sets hl=, so a US-English overview and a German one name different products from different pages — and sometimes one market shows an overview where the other shows none. Keep a separate log per market rather than one file you can't disentangle later.

Summary

For a one-off answer to "does the overview recommend us?", search it yourself and read the sentence. When the question has to be asked every month, across the queries that actually drive revenue, and answered with quotes instead of impressions — fetch the records, keep the sentences, and keep the sources next to them. Two rules keep the number honest: the sentence tells you how you're described, not the mere presence of your name; and a query with no overview is missing data, never a loss.