AdCopilot docsby Atromx
Tool reference

Analysis and audit tools

The read-only analysis tools, from the full audit and account health score to change history and conversion-gap lenses. They quote what they measured.

View as Markdown
Open inClaudePerplexity

These tools are AdCopilot's own, not part of Google's server. Each reads the account, runs the same rules with the same thresholds every time, and returns findings with the numbers it measured and a next step: the call that would fix it where AdCopilot has one; a draft of that call where the fix needs something only you can give, such as the words to block, the text an ad says or the campaign to take budget from, with what it needs named; otherwise the steps in Google Ads. None of them changes anything: your assistant proposes each call to you first.

What each check measures is in Audit checks.

What each call counts

On the Free plan a call counts toward the daily operations by how much it does. full_audit counts 4, at either depth, and so does account_health_score. A deep full_audit narrowed with sections counts by the areas it covers: one area counts 1, two or three count 2, four count 3, and five or six count 4. A full audit or an account health score counts 1 when its query budget cuts it short, or when Google's daily limit on Google Ads requests is reached and it is answered from cached data; one that reuses recent cached reads before the limit is reached counts 4. Seven of the nine lenses, explain_conversion_gap and paid_vs_organic count 2, and so does list_audiences (see Read and report). bidding_audit, day_of_week and change_history are one Google read each and count 1, as every other tool does, shopping_product_issues and lsa_leads among them. On Starter, Plus, Scale and Pro, and during a running trial, every call counts 1; a trial held to Free's 10 a day counts as Free does (see The trial). A call that fails counts nothing, and so does an audit or an analysis that ran out of time and returned only what finished, or a paid_vs_organic call with no Search Console connected. A full audit asked for with no depth whose deep run did not finish in time, so that the quick audit answered, counts 4, as a quick audit does. On Free the result of every call that counts says how many operations it used. See Plan caps and limits.

full_audit

Read. Audits conversion tracking, wasted spend, structure, keywords, ads and settings. Findings come failures first, then by each check's severity, not by the amount measured. Some carry a fixing call: exact-match negatives for searches and close variants that spent with no conversion (never one of your own keywords), a budget raise for each top converter its budget holds back, a pause for a disapproved ad you won't fix. Some carry a draft that names what you supply: the searches to block after a search-terms review, the callouts or snippet values to show, the image, the campaign to move budget from, the keywords and the ad for a new ad group. The rest name the steps in Google Ads. When a result would be too long to send, the drafts are left out first, then the pauses for ads you won't fix, before any finding is.

  • depth="deep" (the default) attempts all 75, scores each one the account has data for, and gives a 0-100 health score, graded over the checks that apply to the account: 73 of the 75 count toward it (the two naming checks are advice only), and a check that does not apply, such as the Performance Max checks on an account with none, is left out of both sides. The six Quality Score rules come back as one finding, each rule with its own verdict and sentence inside it; quality_score_breakdown gives each rule's own table. sections=[…] narrows a deep audit to some of the six areas. If a deep audit asked for without a depth cannot finish, the quick one runs instead and the result says the deep audit was not available this time. On Free, a call whose sections name four or fewer areas, and so count less than a full audit, runs as if asked for with depth="deep" and never falls back: the quick audit ignores sections and counts 4. With five or six areas the call counts 4 either way and falls back like any call that names no depth.
  • depth="quick" scores three of the 75 checks over this month to date, whatever days says, and does not grade the account. days and sections shape a deep audit only.
  • If Google is slow, a deep audit asked for with depth="deep" stops waiting after 45 seconds and returns what finished; it never falls back to the quick audit. A call that names no depth stops waiting at 27 seconds and returns what finished the same way, and only if the deep audit has returned nothing by 36 seconds does it return the quick audit instead, saying so. Checks that did not get their data are listed as not evaluated, never scored. See An audit returned partial results.
  • A check that finds too little or nothing to measure is not evaluated and is left out of the score, and its row in the not-evaluated list says why. On a new account that covers keywords Google has not given a Quality Score yet, a click-through rate on fewer than 1,000 impressions, and anything ranked by spend before anything has spent. On any account it also covers a check whose subject is not set up, such as the conversion checks when no primary conversion action is enabled. The audit says in one line how many checks that was.
  • The audit grades what is running. A removed campaign, and everything in it, is left out of the checks on campaigns, ad groups, keywords, search terms, ads, assets, audiences and targeting, and so are a removed ad group's keywords, ads and audiences. What a removed ad group or keyword did in the period still counts in its campaign's own figures, such as its search terms, and its clicks and impressions in Click-through rate (G-AD2). A few checks take a figure for the whole account rather than campaign by campaign, so what a campaign, ad group or keyword paused or removed during the period did still counts there: the conversion counts in Double counting (G-CT1) and Conversion tag firing (G-CT3), the conversion values in Conversion values (G49), the clicks G-CT3 reads, the change history in Search-term reviews (G13), and the landing pages that Landing page speed for real visitors (G-LP1) picks by their clicks. Paused keywords, ad groups and campaigns are left out of the checks on keywords, Quality Score, ads, assets, audiences and account structure; money a paused campaign spent in the period still counts where a check measures that spend, such as wasted searches, and what a paused ad group or keyword did in the period still counts in its campaign's own figures, such as its search terms and Click-through rate (G-AD2). The Quality Score checks also leave out keywords with no impressions in the period, and the Quality Score finding says how many it left out.

Run a deep audit of my Google Ads account.

account_health_score

Read. One number: the weighted 0-100 health score, graded over those of the 75 checks that apply to the account, its letter grade, the score per category, and how many checks could be evaluated. No findings. If fewer than half the checks that apply could be evaluated, there is no score, and it says why.

The nine lenses

ToolWhat it answers
analyze_wasteWhere money goes with nothing to show: zero-converting search terms and keywords, broad match on manual bidding, campaigns under 1.0 ROAS.
analyze_search_termsNegative-keyword candidates by theme (information, jobs, free, competitors, your never-show-for list, searches in another language), with one ready-to-run add_negative_keywords call per theme. Searches in another language get a call only when a profile language is saved.
keyword_opportunitiesSearch terms that convert but are not keywords, converting keywords losing impression share to rank, and negatives that block a converting term.
quality_score_breakdownQuality Score split into expected CTR, ad relevance and landing-page experience, weighted by impressions, with the worst high-spend keywords named.
rsa_asset_reportEvery responsive search ad headline and description with Google's BEST / GOOD / LOW label, ad strength, counts, pinning and creative age.
conversion_setup_auditWhether conversion tracking can be trusted: primary actions, enhanced conversions, attribution, values, duplicate counting, GA4 link, consent mode, tag firing.
bidding_auditBid strategies against the account's own conversion volume: manual bidding with enough data, unrealistic targets, campaigns stuck learning.
budget_pacingThis month's projected spend against each daily budget, in the account's time zone. A budget several campaigns share is projected once, on their combined spend. month is YYYY-MM.
day_of_week90 days of cost and conversions by day and hour, naming the day-parts that lose money and the ones that are underserved.

change_history

Read. What changed in the account over the last days (1 to 30, default 7, today included, in the account's own time zone): when, by which Google login, and through which channel — AdCopilot, the Google Ads website, auto-applied recommendations, the Recommendations page, a script, an automated rule, a bulk upload, the mobile app, Search Ads 360, or another tool using the Google Ads API. Changes made at the same minute, by the same login, through the same channel, on the same campaign are listed together as one change set, newest first, with each change in a plain line and money in the account's currency. campaign_id narrows it to one campaign.

  • A change counts as AdCopilot's when it came through the Google Ads API within 5 minutes before (or 1 minute after) a change AdCopilot recorded on that account. Another tool writing through the API in the same minutes would be counted with it, and the result says so. If AdCopilot's records cannot be read, those changes are not split between AdCopilot and other tools, except one on the exact item a decision saved in AdCopilot that day names; the result says which rule it used.
  • What Google keeps sets the limits: 30 days of detail, changes made in Google Ads Editor are not included, some rows of the Change History in Google Ads are not available to it, and a change can take about 3 minutes to appear. A call within a minute of an earlier one may repeat its list. Asking for more than 30 days is refused.
  • Two observations come with it: changes made outside AdCopilot, and changes Google applied through auto-applied recommendations. The second names list_auto_apply_subscriptions, which shows which kinds are on, and pause_auto_apply_subscription, which pauses one when you ask.

What changed in my Google Ads account this week, and who changed it?

explain_conversion_gap

Read. Why Google Ads and Analytics report different conversion counts for the same dates. Over the 30 days ending yesterday by default (at most 90), it reads each Google Ads conversion action's window, counting and attribution, its results by click date and by the date they happened, and, when Analytics is connected, the linked property's attribution model, lookback windows and export scope, its time zone, and the key events it credits to this Google Ads account. Both counts are right for what they count.

  • The causes it can measure are sized: the two date bases, results of actions Analytics cannot see (calls, uploads, store visits), and, as an upper bound, credit Analytics gives to other channels. Counting, windows, the export scope, the time zone and the most recent days are named without a size, and what is left is stated as unexplained, never filled in.
  • When several Analytics properties are linked, or a Google Ads tag action has no obvious Analytics key event, it asks: ga4_property and ga4_event carry the answer back. conversion_action_id narrows it to one action.
  • When Analytics withholds low counts, its figures are given as lower bounds.
  • It proposes no change. It names the settings either side could change and what each would do; ga4_update_attribution_settings changes Analytics' model or windows only when you ask.

Google Ads says 14 sign-ups last month and Analytics says 9. Why?

Read. Sets the search terms the account pays for in Google Ads beside the queries a Search Console site already ranks for, over the same dates. By default that is the 28 days ending three days ago, because Search Console lags, and at most 90. It needs Search Console connected: without it the call is refused with how to connect it, and is not counted as an operation. When several sites are connected and none is named, it asks which with site_url.

  • It lists terms paid for while ranking in the top three, paid terms ranking lower, queries with 20 or more impressions that no Google Ads search term was found for (split between ranking in the top three and ranking fourth to twentieth), and paid terms with no organic row. Each carries cost, clicks, conversions and average position where the account has them.
  • A side-by-side of the top paid terms by cost and the top organic queries by clicks always comes back, even when nothing overlaps.
  • Terms are matched on identical wording only, after lower-casing and ignoring punctuation: "shoe" is not matched to "shoes". Brand terms are marked, left out of the counts and costs of the terms to review, and not offered as something to review, because whether to pay for your own name is a different decision.
  • Search Console withholds rare queries and runs about three days behind, and position is an average. "Paid only" means no organic row was found, "no Google Ads search term found" means none in the search-term report (Google Ads hides low-volume terms and the report leaves out Performance Max terms, so the query may already have an ad), and when Search Console returns its full 5,000 rows that list is labelled as not covering every query.
  • It proposes no change; it names terms to review or test by hand.

Which of the terms I pay for does my site already rank for?

shopping_product_issues

Read. Which of your Merchant Center products cannot show in Shopping and Performance Max ads, and why, from Google Ads' own view of the products your linked Merchant Center accounts send it. It needs no new sign-in: it reads with the Google Ads access you already gave.

  • Counts by status: products that can show, products that can show only in some places, and products that cannot show.
  • The issues behind them, grouped by Google's error code and the attribute it names, the ones on the most products that cannot show first, each with Google's own description, its fix text, its Help Center link and the regions it affects (none listed means every region), with a few item ids as examples.
  • The products that cannot show, the costliest first (25 at most; max_items asks for fewer), with the clicks and cost each had over the last 30 days, so you can see what stopped serving.
  • It reads the products each enabled Shopping or Performance Max campaign includes, on the two that spent the most (it names the others), so a Merchant Center product no campaign advertises is never counted as one that cannot show. A product counts as unable to show only when it cannot show in any campaign that includes it.
  • campaign_id narrows it to one Shopping or Performance Max campaign, and the status and numbers are then that campaign's, as Google reports them. An account with no enabled Shopping or Performance Max campaign, or a paused campaign asked for by campaign_id, is told so without a read.
  • Two audit checks come with it: the share of products that cannot show, and whether one issue stops most of them. Both are in the deep audit too.
  • Google can take up to a day to update a product's status, after a fix as well. It proposes no change: a product is fixed in Merchant Center (or, if Google says a listing group leaves it out, in the campaign), and AdCopilot does not write there.

Why did my Shopping ads stop showing some of my products?

lsa_leads

Read. Your Local Services ads' leads over a window, and the business's verification, from Google Ads' own Local Services data. It needs no new sign-in: it reads with the Google Ads access you already gave. See Local Services for the whole read and for rating a lead.

How did my Local Services leads go this month, and which still need a rating?

Output

Each tool takes detail="summary" or "full" and render="json", "markdown" or "both". Every result carries a coverage block that names the checks not evaluated and why. Read it before reading a result as a clean bill of health.

While Google Ads requests are paused, the tools answer from cached data and say so; fresh audits resume when the pause ends.

On this page