# Changelog of Google AI Overview Tracker: Brand Mentions & Citations (`fetchwright/google-ai-overviews-ai-mode-scraper`) Actor

- **URL**: https://apify.com/fetchwright/google-ai-overviews-ai-mode-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/fetchwright/google-ai-overviews-ai-mode-scraper.md

## Changelog

### 2026-11-16

- README: a short note and Apify's video on running it on a schedule.
- README: related actors updated (same-topic links first, current names).

### 2026-11-15

- README first check: a fourth step to schedule the check weekly with Compare with the previous run (about $0.47 a month on Free, $0.29 on Business for 10 keywords x 3 searches).
- README links to ready-made brand-visibility scripts on GitHub.
- Added `listedNames` to AI Overview and AI Mode samples: the short labels that start the answer's list items (before a colon), including products you did not track; ranking labels ("Best for ...", "Top pick") and meta labels (Pros, Pricing) are left out, and for "how to" searches the labels are often criteria rather than products. AI Overview query summaries include up to 20 names with answered-sample counts, shares and 95% intervals. Names are unverified; check before using as brand data.
- FAQ: "Is there a Google AI Overview API?" Google offers no API that returns the AI Overview text or its cited pages; this Actor works as one.
- README opens with a two-minute first check: the example task to start from, a typical cost (10 keywords x 3 searches: about $0.12 on Free, $0.07 on Business, add-ons off) and an abridged real output row. The pricing-update note moved here: first-view measurements are an optional add-on on paying plans at $0.90 per 1,000 searches and included on Free (14 November 2026).

### Unreleased - 2026-11-14

- First-view checks match cited pages by exact normalized URL and passages by complete words. A check needs a decided passage or cited-page flag for the add-on fee; unmatched observations include an explanation.

- Paying-plan searches and optional measurements stop before a request would leave too little time to save the result. When a re-search cannot fit, the earlier result keeps its observed status and the summary reports the time limit.

- Searches and optional measurements reserve room under the maximum charge before starting, including requests already in progress. Add-ons that do not fit are skipped.

- Paying-plan searches include places and top stories when supplied; the page-layout classification remains null. Missing or pending AI Overviews receive the configured re-searches, with one search fee per sample.

- A CAPTCHA or refusal disables the search fallback for the run; primary searches can continue. The run summary describes which route was refused.

- Optional first-view measurements on paying plans: $0.90 per 1,000 searches with a measurement; included on Free. Enable `firstViewCheck` to get first-view text, flags for matched cited pages and tracked brands, and source shares in the summary. Unavailable checks stay null with reason "first-view unavailable", have no add-on charge and leave search results available.

- Lower prices per 1,000, Free / Bronze / Silver / Gold: search page $3.00 / $2.70 / $2.40 / $2.05; AI Overview returned $0.90 / $0.60 / $0.50 / $0.40; AI Mode answer $5.00 / $4.00 / $2.50 / $1.75. Platinum and Diamond use Gold prices.

- Searches with an AI Overview now total $3.90 / $3.30 / $2.90 / $2.45 per 1,000. README job estimates and the comparison with apify/google-search-scraper are updated (checked 14 November 2026).

- A paying-plan run stops starting new searches when the primary route becomes unavailable and ends with partial status "primary route unavailable". Saved results remain available; searches not done are not charged. Free-plan runs continue to use the standard route.

- Busy searches wait and retry up to three times without switching to the backup. Paying-plan runs allow up to 5 simultaneous primary searches by default, within a shared account limit of 25.

### 0.1.81 - 2026-11-14

- One name everywhere: the listing, README and input form now call this Actor Google AI Overview Tracker, and the listing and README opening state the Free-plan price and a paid-plan price; the README opening also explains the required keywords and optional brands. The README now also names the forum threads and related searches returned when Google shows them on English-language result pages. No change in behaviour or price.

### 0.1.80 - 2026-11-13

- README: the "Update 10 November 2026" box below the opening now says, in plain words, which questions the Actor answers. No change to the Actor, its input or its output.

### 0.1.79 - 2026-11-13

- README: a "Use it in a pipeline" section with one step-by-step recipe for Make (weekly AI Overview rates into Google Sheets), using Apify's own integration and your own Apify account. It says what it was checked against and that it was not run in a live account. New section with Python, Node.js and curl examples for the API, and a FAQ answer on downloading the results as CSV, Excel or Google Sheets. No change to the Actor, its input or its output.

### 0.1.78 - 2026-11-13

- Wording: a change flagged by **Compare with the previous run** is now described as what it is: the two runs' rates differ by more than their 95% intervals allow, for these samples at that time, country and language. Samples in one run can move together, so the flag is a change indicator, not a statistical test. The field names (`significant`, `...Changed`) and the rule itself are unchanged.

### 0.1.77 - 2026-11-13

- Reliability: a video's source line is tidied in one pass, so a result whose text holds a very long run of line breaks can no longer slow a run down. Results are unchanged.

### 0.1.76 - 2026-11-13

- Google's verification form is now recognised by reading its tag as a browser does: a ">" inside a quoted attribute no longer hides the form, character references with leading zeros are decoded, and the wider forms added in 0.1.75 (spaces around "=", other quotes) now count only inside a real form tag, not inside another attribute's value, a comment or script content, and an id with spaces inside its quotes (" captcha-form ") no longer counts, as a browser would not match it either. The original check (id="captcha-form" written exactly so) is kept as it was, so every page recognised before 0.1.75 is still recognised.

### 0.1.75 - 2026-11-13

- Reliability: the check for Google's verification page now takes time in proportion to the page's length; a page with many unclosed links could make it take seconds or minutes before.
- Google's verification form is now also recognised when its id is written with spaces around "=", with single quotes or none, or with character references (for example `id = "captcha-form"`). Before, such a page that also showed a link could be saved and charged as a search; now the run stops searching, as for any verification page.

### 0.1.74 - 2026-11-12

- README and input help: the AI Mode answer rate now includes the 12 November 2026 test (14 of 20), and the AI Mode timing is the measured one (20 queries in 702 seconds; one slow answer can take over two minutes) instead of "15 to 60 seconds per query".

### 0.1.73 - 2026-11-12

- A search whose AI Overview Google was still preparing is now named in the run's status message (for example "0 AI Overviews, 1 still loading when Google answered (AI Overview fee not charged)") and counted in the run summary as `totals.aiOverviewsStillLoading`, instead of reading only as "0 AI Overviews". Charges are unchanged: the search page is charged, the AI Overview fee is not.

### 0.1.72 - 2026-11-12

- Each search now also returns the Top stories block when Google shows it (title, publisher, age and the article address when the card links to it directly), at no extra fee.

### 0.1.71 - 2026-11-12

- Each search now also returns the Places block (business name, rating, review count, price range, category and address line as shown; never review quotes or phone numbers) and the Videos block (title, source, age, duration and the video address when shown), at no extra fee.

### 0.1.70 - 2026-11-12

- Each search now also returns the related searches Google suggests and the threads it shows under Discussions and forums (title, address, site, community, comment count, age), at no extra fee.

### 0.1.69 - 2026-11-12

- README: how long a run with AI Mode takes (measured on 20 queries), the longest one AI Mode answer can take, and that a run ends with the rows it saved before its timeout.

### 0.1.68 - 2026-11-12

- README: a new answer lists the field names of other AI Overview actors that this Actor reads, and when its own field gives way to them.

### 0.1.67 - 2026-11-12

- When Google answers an organic-address lookup with a verification page, a refusal or a rate limit, no further address is looked up in that run, by any route; addresses are also not looked up after a search was refused.
- A run that the platform restarts no longer searches and charges a second time: it stops, keeps what was saved and says so.
- A sample's row is saved before it is charged: a row that cannot be saved is not charged, and samples running at the same time can no longer go past your maximum charge together.
- If a charge reports an error after its row was saved, the run makes no further search and the `SUMMARY` record lists that row and event under `chargeErrors`, because whether the platform recorded the charge is not known.
- A search that Google refuses (HTTP 403) now ends the run's searches, as a verification page already did; a plain rate limit is still tried again after a pause.

### 0.1.66 - 2026-11-12

- README: the cost examples for 150 and 600 searches are corrected to the cent ($0.68 and $2.74, not about $0.70 and $2.80).

### 0.1.65 - 2026-11-12

- Searches whose results or answer talk about Google's "unusual traffic" message are read like any other search; before, such a search could be reported as failed (not charged). The first part of this went out in 0.1.64.

### 0.1.63 - 2026-11-12

- A run that stops because Google asked for verification now says so in the run summary (`stoppedBecause`). A page that only mentions such a message in an answer or a result is no longer mistaken for one.

### 0.1.62 - 2026-11-11

- If Google answers a search with a verification page, the run stops searching at once and does nothing to get past it. The searches not made are reported as failed with that reason and are not charged.

### 0.1.61 - 2026-11-11

- Client reports name one more kind of offer found in ad wording: free cancellation.

### 0.1.60 - 2026-11-11

- The README now shows a real measurement from 11 November 2026 (three queries, five searches each, six brands) with its cost, its intervals and a downloadable file, so you can see what the numbers look like before running.

### 0.1.59 - 2026-11-11

- Input written for other AI-visibility actors now works as it is: `prompts` and `customQueries` are read as queries, `runsPerPrompt` as samples per query, `brandName`, `brandAliases`, `brandDomain` and `competitors` as the brand list, `includeReport` as the report switch and `webhookUrl` as the alert address; the log says how. A field given under this Actor's own name wins.

### 0.1.58 - 2026-11-11

- The run's storage now describes its records (the run summary, the client report), so they show as named tabs in the run's Storage view, and the main input's help says it is required unless a dataset is used. No change to results or prices.

### 0.1.57 - 2026-11-11

- The rates you run this Actor for now arrive in the dataset itself, not only in the run summary: after the samples, the dataset ends with one extra row per query you asked for, with that query's AI Overview rate and its range, first-try versus eventual rate, each brand's mention and citation rates with their ranges, the cited sources, the outcome (answered, no AI Overview shown, still being prepared, failed, not searched) with its counts, when the searches were made and, with "Compare with the previous run", what changed. API, MCP, Zapier, Make and spreadsheet users get them without opening the run summary. A new dataset view, "Results per query", shows them as a table.
- Every row now has `rowType`: `sample` for a search, `query_summary` for the new per-query rows. Anything that counts or reads rows as searches should filter on `rowType: "sample"`. The new rows are free; charges, the run summary, alerts and the comparison with the previous run are unchanged. A dataset from this Actor used as the query list still gives each query once.

### 0.1.56 - 2026-11-11

- The shared report engine was updated (an offers section for Google ads reports). No change to AI Overview reports, results or prices.

### 0.1.55 - 2026-11-11

- Listing text updated. No change to results or prices.

### 0.1.54 - 2026-11-11

- An alert that fails to deliver is tried up to three times, and the run summary says whether it was sent (`notification.deliveryStatus`) and after how many tries. The results are saved either way.

### 0.1.53 - 2026-11-11

- The listing links a corrected sample report and a second one with a query Google did not answer.

### 0.1.52 - 2026-11-11

- The client report's count of sites cited in every answer now always matches the list beside it and the data (the sample report said 8 where its own list had 9).
- A quoted answer is now cut around the place where the brand is named, however far into the answer that is. The report no longer says the text ends before the brand when the brand is in the answer.
- The report and the run summary now give, per query, the real time from the first to the last search, the number of searches saved and how many different answer texts they returned, instead of saying the searches were minutes apart. The report also states which version of the Actor collected the data.
- New: every query you asked for is listed in the run summary (`coverage`) and in a closing table of the report with its outcome: answered, no AI Overview shown, still being prepared, failed, not searched because the run reached its maximum charge or time limit, or too few answers for statistics. A run stopped by its maximum charge now gets a report that says which queries were not searched.
- Compare with the previous run: a run in which Google showed no AI Overview is now the new baseline, and the summary and alert say once when the AI Overview disappears (`lost`) and once when it comes back (`recovered`). Runs whose searches all failed or stayed pending still leave the baseline alone. A previous record saved before version 0.1.34, which has no settings, is no longer compared: the run says comparability is unknown and starts a new baseline.
- The input help, README, report and alert now describe a marked change the same way: the two 95% intervals do not overlap, a rule of thumb rather than a statistical test.
- A partial AI Overview now counts toward the three answers a query needs to appear in the report, as it already did in the run summary.

### 0.1.51 - 2026-11-11

- The listing's first lines now say which parts are options: repeated searches, AI Mode and the comparison with the last run. (0.1.50 was a rebuild with no change.)

### 0.1.49 - 2026-11-11

- The listing's first lines now start with what you get: whether Google's AI answers mention your brand, who they name instead and which pages they cite. No change to results or prices.

### 0.1.48 - 2026-11-11

- The listing now has a dated feature comparison with the four most-used AI Overview actors, read from their public listings, including where they do more.

### 0.1.47 - 2026-11-11

- A query list pasted with bullets, numbering (1., 2., 3. on three or more queries) or the quotes and commas of a JSON list is now searched without them, and the log says what was removed. Several queries pasted as one entry on separate lines are searched separately. Before, the marker was sent to Google as part of the search. A query that is only wrapped in quotes is still searched as an exact phrase, and the log now says so.

### 0.1.45 - 2026-11-11

- Several brands pasted on a single line ("HubSpot, Pipedrive, monday.com") are now tracked as separate brands. A company ending after a comma ("Hubspot, Inc.") and names such as "Goodby, Silverstein & Partners" stay whole; in a list with several entries each entry is taken as typed. A leading @ is ignored.
- A brand too short to look for (a single character) now gets a warning in the log instead of silently showing as never named.

### 0.1.44 - 2026-11-10

- The Output tab's client report link now always opens a page: the report when one was written, otherwise a short note saying why there is none and what to change, with a link to a sample report.

### 0.1.42 - 2026-11-10

- A long run now reports its progress every few seconds in the status message and the log: searches done of those planned, and how many returned an AI Overview so far.
- The client report for runs with no own brand marked now leads with the most notable finding, and treats every brand the same.

### 0.1.39 - 2026-11-10

- New: a client-ready report, free. Switch on "Write a client-ready report" and the run also saves REPORT.html and REPORT.md: the strongest finding about the tracked brands, how often each is named and cited per query, who Google cites instead, and quoted answers as proof. Every figure is counted from the run's rows; no AI model writes it. It never changes the run's results or charges.

### 0.1.35 - 2026-11-10

- New: alerts. Paste a Slack or Discord incoming-webhook address (or any HTTPS address) and each run posts a short message: AI Overview counts per query, how often each tracked brand was named and cited, and, when comparing with the previous run, the sources that appeared or disappeared and the changes marked significant. Optionally only when something changed.

### 0.1.34 - 2026-11-10

- The run summary now lists, for each query, the samples requested, the samples that failed (not charged), the samples still pending and the samples decided, and every rate gives the count it is divided by.
- New: how often the first search of a sample showed an AI Overview, with its 95% interval, kept apart from the result after free re-searches, and the total number of searches made.
- Compare with the previous run: shows when the previous run was made and how many hours ago; runs made with different settings are marked as not comparable, with the differing settings named and no change marked significant; a query with no decided samples in this run no longer reports sources or brands as lost.
- Documentation: the price comparison now also covers paid plans, where Apify's own scrapers cost less for the raw answer.
- Documentation: the significance rule is stated exactly (the two 95% intervals do not overlap) and described as a rule of thumb; "before Show more" is described as a property of the page, not of what people saw.

### 0.1.33 - 2026-11-10

- Documentation: a dated price comparison.

### 0.1.32 - 2026-11-10

- Cited sources now include the sites behind a "+2"-style citation chip (for example a forum cited only there); they are listed by site, with `url` null. Source counts, source types and agreement figures use them.
- The answer text and blocks no longer contain the caption of a video card embedded in the answer.
- With several queries, a query whose AI Overview stays unfinished is re-searched less often.
- Documentation: the default number of re-searches is 1.

### 0.1.31 - 2026-11-10

- New: **Compare with the previous run**. For scheduled tracking, the run summary says per query which sites entered or left the sources and how each site's share and each brand's mention and citation rate changed, marking a change significant only when it is unlikely to be chance.

### 0.1.30 - 2026-11-10

- Documentation: AI Mode answer rate measured on a larger sample (18 of 20 queries on 10 November 2026).

### 0.1.29 - 2026-11-10

- New images in the documentation.

### 0.1.28 - 2026-11-10

- New: `aiOverview.blocks`, the answer as ordered blocks (heading, paragraph, list item, table row), each with the sites and pages cited for that passage and whether it is shown before "Show more".
- Documentation rewritten for clarity: who it is for, why choose it, use cases by role.

### 0.1.27 - 2026-11-10

- Documentation: images and speed figures updated with output measured on 10 November 2026.

### 0.1.26 - 2026-11-10

- When Google is still preparing an AI Overview, the Actor searches once more by default (you can set 0 to 3). A query that stays `pending` is not searched again for its remaining samples in the same run, so runs finish sooner.
- Documentation: which queries and countries tend to return `pending`, with measured examples.

### 0.1.25 - 2026-11-10

- New: for every cited page and tracked brand, whether Google shows it before "Show more" (`inFirstView`), and the answer text shown first (`textFirstView`). The summary gives each source's and brand's first-view share.
- New: a 95% confidence interval and the number of samples on every rate in the summary.
- New: with the AI Mode add-on, AI Overview vs AI Mode for each sample and query: sites cited by both, by only one, and an overlap score.
- New: answer-text agreement between samples (`textAgreement`), next to source agreement.
- New: each AI Overview source's position in the organic results of the same search, and per query the share of citations from sites that do not rank there.
- New: each cited site is classed as user content, government or education, your brand, a competitor, or other, with shares per query.
- Brand tracking: entries can now be objects with aliases, domains and a competitor flag (plain names still work). New fields: how often each brand is named, the order in which brands are named, share of voice. A name typed with a capital letter no longer matches the ordinary word, and a brand counts as cited when any of its domains is cited.
- New status `pending`: Google was still preparing the AI Overview when the page was returned. Such searches were previously reported as `not_available`. They are never charged the AI Overview fee, are listed under `pendingSamples` in the summary, and are left out of the AI Overview rate.
- The free re-search now also covers searches where Google is still preparing the AI Overview, and searches up to twice by default (new option: Re-searches for a missing AI Overview, 0 to 3).
- Tables in answers keep their rows: one row per line, cells separated by " | ".
- AI Mode replies end with the answer: copy, share and feedback labels are no longer included.
- Organic results: the snippet no longer starts with the site name and date; the date has its own `date` field.
- Summary: cited sites and publisher names (sources shown without an address) are listed separately, and source agreement compares sites only.
- Country `UK` is accepted as `GB`, and an unknown country code stops the run at once with an example. Searches now ask for the chosen language. The run log warns when a chosen dataset holds more queries than are read.
- Organic results to keep: the maximum is now 10, which is what one results page holds.

### 0.1.24 - 2026-11-10

- Fixed: long AI Overviews were cut off after the part Google shows first. The whole answer is now returned, and the sources and source descriptions from the later part with it. If you sampled long answers before this version, run them again.

### 0.1.23 - 2026-11-10

- Long runs now finish cleanly before the run timeout: every saved row is kept, the run succeeds, and queries with samples left are listed under `unfinishedQueries` in the `SUMMARY` record. The same list is written when a run stops at your maximum charge.

### 0.1.22 - 2026-11-10

- More organic results carry their full address in the responses where Google hides it on the page (11 of 26 in our test). No extra charge. An address that cannot be recovered is left empty, never guessed.

### 0.1.18 - 2026-11-10

- Documentation: the connector address for using this Actor as a tool in Claude, ChatGPT and other AI apps.

### 0.1.17 - 2026-11-10

- New images in the documentation. No change in behaviour.

### 0.1.16 - 2026-11-10

- Maintenance release. No change in behaviour.

### 0.1.14 - 2026-11-10

- New: Read queries from a dataset, to chain this Actor after another one.
- The final status message now says what was not charged.

### 0.1.13 - 2026-11-10

- Two more result views: Cited sources and Brand visibility.
- Input form reordered: samples per query now comes second; clearer field help with examples.
- README: Limits section and three more FAQ answers.

### 0.1.12 - 2026-11-10

- New: when Google reports that no AI Overview is available, the search is repeated once at no extra charge. Rows record `aiOverviewAttempts` and `firstAiOverviewStatus`. Can be turned off.

### 0.1.11 - 2026-11-10

- Cited sources now include the page title and description when Google shows a source card for the page.
- Added this changelog.

### 0.1.9 - 2026-11-09

- First public release.
- AI Overview text with headings and lists kept, cited source URLs and domains, People also ask and organic results from the same search.
- Samples per query (1 to 10) with a run summary: how often an AI Overview appeared, cited sources with their share of samples, source agreement between runs.
- Brand tracking: mentioned in the answer, cited as a source.
- Experimental add-on: Google AI Mode answer, charged only when an answer is returned.
- Explicit AI Overview status (available, partial, not_available, absent). The AI Overview fee is charged only when an answer is returned.
- Cited source URLs are returned in every response format Google uses.
- Never bills past the maximum charge set for a run.
