AI Overview Rewrite Queue | Find Pages Losing Clicks to Google's AI

SEO API that joins your Google Search Console export with live AI Overview citation data and returns a scored, tiered list of which pages to rewrite first. It finds the queries where you already rank but Google's AI is answering with somebody else's content, and tells you which ones are worth the work.

Ranking and being cited are two different problems. Search Console shows you the first one. This shows you both at once.

What this API returns

One row per query, sorted so the highest-leverage rewrites come first:

• Your Search Console data: clicks, impressions, CTR, and average position, parsed from whatever format your export used.
• Live citation data: whether an AI Overview appeared, which of your pages it cited, and every domain it cited instead.
• A tier from A to X with a one-sentence reason in plain language, so the row explains itself without a legend.
• A join label on every row, so a query that appears on only one side is visible rather than missing.
• A whole-run summary in the key-value store: counts per tier, the join rate, and the competitor domains cited most often across your keyword set.

What makes it different: no other tool on the Store puts Search Console position next to AI Overview citation status. Citation-only trackers tell you that you were not cited. This tells you which page to open first, and why.

The five tiers

Tier	What it means	What to do
A	You rank 5 to 20 and a competitor is cited instead of you	Rewrite first. Google already understands the page; someone else answers the question better.
B	You rank 1 to 4 and the AI ignores you, or cites nobody	Answer-shape problem, not an authority problem. Restructure the answer, do not chase links.
C	You are cited and still converting below your own baseline	The overview is satisfying the search. Consider what the page offers beyond the answer.
D	No AI Overview, or nothing indicating a rewrite	Ordinary SEO. Listed so the queue stays complete.
X	The check did not complete, or the query is on only one side of the join	Unknown. Evidence, not a conclusion.

Tier C compares each query against your own median CTR at a similar position, computed from the no-overview queries in your own export. It never uses an industry table, because every account's normal is different.

Example output

One real row, trimmed:

{
  "result_type": "scored_query",
  "query": "what is a crm",
  "tier": "A",
  "tier_reason": "You rank at position 8.4 and a competitor is cited in the AI Overview instead of you, on 1,900 impressions. Google already understands this page; somebody else is answering the question better. Rewrite this one first.",
  "join_status": "matched",
  "clicks": 12,
  "impressions": 1900,
  "ctr": 0.0063,
  "position": 8.4,
  "check_status": "ok",
  "ai_overview_present": true,
  "citation_state": "competitor_cited",
  "cited_urls": [],
  "cited_pages_count": 0,
  "reference_count": 8,
  "reference_domains": ["reddit.com", "youtube.com", "ibm.com", "zendesk.com"],
  "fetched_at": "2026-08-17T22:41:03.118204+00:00"
}

A five-row queue looks like this:

Tier	Query	Position	Impressions	Citation state	Why
A	what is a crm	8.4	1,900	competitor_cited	You rank and a competitor is cited. Rewrite first.
A	crm integrations guide	6.0	4,300	competitor_cited	Same shape, more traffic behind it.
B	crm software, cheap	2.1	320	overview_no_references	You rank top 3 and the overview cites nobody.
C	best crm for small business	4.2	2,600	cited	Cited, converting under your own median at this position.
X	free crm tools	14.2	180	null	The check did not complete, so this is unknown, not an absence.

Use cases

• Build a rewrite backlog from evidence rather than from opinion, by exporting Search Console once a month and running the queue.
• Find the answer-shape problems, the pages that rank in the top 4 and still get passed over by the AI.
• See who is answering your queries: the run summary tallies the domains cited most often across your keyword set.
• Track the three states over time on a schedule, since a single check is noise. The same query can cite you Monday and drop you Thursday.
• Feed an agent: an MCP client can pull the tier A list and draft the rewrites in the same session.

Input parameters

Parameter	Required	Default	Description
target_domains	Yes		Domains you consider yours, for example example.com. Subdomains match their parent.
search_console_csv_url	No		URL of a Search Console Queries export. A published Google Sheet CSV works and lets scheduled runs re-fetch fresh data.
search_console_rows	No		The same rows pasted as JSON objects, instead of a URL.
queries	No		Extra keywords to check on top of the export.
min_impressions	No	10	Ignore export rows below this impression count.
gl	No	us	Two-letter country code for the citation checks.
hl	No	en	Two-letter language code. AI Overviews are currently English-only.
location	No		Optional named location, for example Austin, Texas, United States.

At least one of search_console_csv_url, search_console_rows, or queries is required. No Google sign-in, no OAuth, and no service-account credentials: it reads an export you already have.

How to export from Search Console

1. Open the Performance report in Search Console.
2. Choose your date range, then open the Queries tab.
3. Click Export and pick a format, or export to Google Sheets.
4. For scheduled runs, put the sheet in Google Sheets and use File, Share, Publish to web, then choose comma-separated values. Paste that URL into search_console_csv_url and every run picks up fresh data.

Localized column headers, semicolon delimiters, comma decimals, percent signs on CTR, and non-breaking-space thousands separators are all handled, so an export from any locale works without editing.

How to get started

1. Open the Actor and put your domain in Your domains.
2. Paste your Search Console export URL, or paste rows directly.
3. Run it. The queue arrives sorted, tier A first.

View on Apify Store

Pricing

Pay per event, with no subscription:

• A one-time setup event per run, covering the work before the first row exists: reading and parsing the export, running the citation checks, and computing your CTR baseline.
• A scored row event per query returned with a tier, a reason, and its joined data.

Current rates are on the Store card, which is always authoritative.

One thing to know about cost. The citation checks are performed by the Google AI Overview API, which this Actor runs on your account. Those runs are billed to you separately, under that Actor's own pricing, and they are the larger part of the cost of a run. Retrieval is not reimplemented here, so you pay for it once, at source. Use min_impressions to keep the checked set to queries with enough traffic to matter.

🔌 Use this API from Claude (MCP)

Add this Actor as a tool in Claude Code (free trial), Claude Cowork (free trial), or any other MCP client, through the hosted Apify MCP server:

https://mcp.apify.com/?tools=actors,docs,johnvc/ai-overview-rewrite-queue

Ask it for the tier A list and it can draft the rewrites in the same session.

Apify MCP integration docs: https://docs.apify.com/platform/integrations/mcp

💸 Pay per run with crypto (x402)

The AI Overview Rewrite Queue supports agentic payments via the x402 protocol. AI agents and MCP clients can pay for runs in USDC (on Base) with no Apify account or API token needed: point your agent at the Apify MCP server and it can discover, pay for, and run this Actor autonomously. Read the Apify x402 announcement for details.

Related tools

• Google AI Overview API: the citation checks behind this Actor. Use it directly when you want raw overview text and sources without the Search Console join.
• Brave AI Mode API: the same visibility question on a different AI answer engine.
• Naver AI Overview API: the same citation question for Korean search.
• Google Autocomplete API: find the question-shaped queries that trigger overviews in the first place.

FAQ

How do you tell a real "no AI Overview" from a failed check?

They are different fields and they are never conflated. ai_overview_present: false means the query was checked and no overview appeared. null with check_status of retrieval_failed or blocked means it could not be checked. Failed checks land in tier X and can never be scored as a rewrite priority, because a failure that reads as a zero is worse than no data at all.

Why is my join rate only 40 percent?

Search Console anonymizes low-volume queries, so long-tail keyword lists commonly match somewhere between 30 and 60 percent. That is normal and not a fault in the export or the Actor. Unmatched rows are kept and labelled gsc_only or check_only, never dropped, so you can see exactly what did not line up.

Which of my pages is being cited, not just whether my domain is?

cited_urls lists the exact reference links belonging to your domains, and cited_pages_count counts them. reference_domains lists every domain the overview cited for that query, in order.

Are AI Overview citations correlated with classic top-10 rankings?

Not reliably, which is the reason this Actor exists. Tier B is precisely the set of queries where you rank in the top 4 and the AI still passes you over. If citation tracked ranking, tier B would always be empty.

How do you handle AI Overview volatility?

A single check is a snapshot, and the same query can cite you one day and drop you the next. Run it on a schedule and trend the tiers rather than reading one run as a verdict. A published Sheet URL in search_console_csv_url makes scheduled runs pick up fresh Search Console data automatically.

Can I pin the country or region?

Yes, with gl, hl, and location, and you should always set them deliberately. Checks run from datacenter addresses, so an unrecorded region makes a lost citation indistinguishable from a different exit node. The region actually used is recorded in the run summary.

Who else is being cited for my queries?

The run summary in the key-value store tallies top_competitor_domains across the whole run, ranked by how many of your queries each domain was cited on.

What if my export is from a non-English locale?

It will parse. Semicolon delimiters, comma decimals, percent signs, byte-order marks, non-breaking-space thousands separators, and localized column headers are all handled. Rows that genuinely cannot be read are counted in the run summary rather than silently discarded.

Does the tool distinguish being mentioned from being recommended?

Not yet. It reports citation, which is a link to your domain in the overview's sources. Whether the surrounding text recommends you, merely mentions you, or contradicts you is a text-analysis question this Actor does not answer.

Why do I see charges from a second Actor?

The citation checks run on the Google AI Overview API under your account, so they appear as their own runs and their own charges. Retrieval lives in one place instead of being duplicated here, which is why it is billed once, at source.

---

🌐 About Alpha OSINT

This Actor is part of Alpha OSINT, toolset of financial and operations data sources and APIs. See the AI Overview Rewrite Queue source page for related tools and use cases. For support or requests for this actor, please start a ticket directly on our support page.

Last Updated: 2026.08.17