AI Visibility Tools vs. Agencies vs. Done-for-You: Who Actually Ships the Fix?
October 4, 2026
•
12
min read
Say you have paid a $4,000 monthly retainer for six months and your cost per acquisition has not budged. The monthly deck still celebrates impression share and click-through rate, while qualified pipeline sits where it did two quarters ago. That is not a messaging disagreement. It is an execution question, and the change log is where I would start.
When you ask what is happening, you may hear about market conditions, aggressive competitors, or the Google Ads learning phase. Any of those could matter. None tells you what the agency actually did. Before you replace the team, renew its contract, or argue through another slide deck, pull the account record.
This is a 30-day test, not a verdict based on a screenshot. You will compare activity across three periods, sort logged changes by commercial consequence, and check the results against what happened to your business. My expectation: when performance has stalled and the log contains little beyond automated or cosmetic activity, the agency will struggle to show how its management earns the retainer. The mechanism is simple. A report can describe work; a change history can show which account edits happened. It cannot show every hour of thinking, so keep that limit in the test.
You need standard or administrative access to your Google Ads account. If the agency controls the account and will not give you direct access to its data, resolve that first. You cannot run an independent account audit from a PDF the agency chose to send you.
In the desktop interface, open Campaigns and select Change History. Google's native interface lets you review a rolling two-year record. Use the date and user filters, inspect the details of individual changes, and export the periods you need. The User field helps distinguish a named operator from the Google Ads system or another tool. For a change whose context is unclear, use the “Go to…” control to inspect the affected campaign or ad group.
Set the test period to 30 completed days ending two days ago. Then pull two comparison windows: the 30 days immediately before it and the first 30 days of the agency's engagement. Use the same account and the same classification rules for all three. Onboarding usually involves building or restructuring campaigns, so I expect more changes there than in a mature account. I do not expect an onboarding count to serve as a permanent monthly quota.
Before scoring anything, write down the business baseline for each window: spend, CPA, qualified pipeline or closed-won revenue if you have it, and any material change in offer, budget, landing page, or conversion tracking. You are testing management against outcomes, not assuming that every performance shift came from an account edit. Excluding the most recent two days also gives conversions some time to appear; use the same cutoff when comparing performance.
Keep a second boundary in view: the change log is not a complete diary of work. It will not capture every investigation, decision not to edit, or change made outside Google Ads. Conversion-tracking work may require a separate inspection, particularly when it happens in Google Tag Manager. Our 10-point Google Ads audit framework treats conversion integrity as its own checkpoint. For this test, score the account changes you can see, then ask for evidence of consequential work you cannot.
A raw row count is a bad score. One bulk action can generate many entries; automated changes can fill a month; and a renamed ad group does not become strategy because it has a timestamp. Export the log and give each relevant entry one of three labels:
Do not infer automation from a crowded timestamp alone. Bulk edits can be deliberate. Open a sample of the entries, check the user or tool, and ask what decision drove them. Conversely, an agency does not get to call auto-applied recommendations hands-on optimization merely because it enabled them. Unmonitored rules can make consequential changes; the useful question is who reviewed those changes against your economics.
For ambiguous rows, apply one control consistently: what commercial variable could this edit change? If you cannot identify an effect on who sees the ad, what the auction costs, what the visitor reads, or where the visitor lands, put it in the clerical column. Separating vanity updates from structural work matters more than arguing over whether the account looks busy.
For every high-impact entry, record its date, campaign, old and new state, apparent purpose, and any supporting performance evidence. Do not award a point because a change sounds technical. An edit counts as a meaningful intervention when you can connect it to a specific problem the agency was trying to address. I would look in four places.
Do not turn those four buckets into a quota. Account size, campaign type, and business changes affect which interventions make sense. The expectation I would test is narrower: if CPA has stalled and the agency says it is actively improving the account, you should be able to find decisions aimed at that problem or credible evidence of work outside the log. If all you find is clerical movement and automatic recommendations, the activity count is doing the selling instead of the work.
Count the high-impact interventions in your test window and compare that count with both earlier windows. If a $4,000 retainer bought four identifiable interventions, that is $1,000 per logged intervention. It is a prompt for a hard conversation, not an hourly rate: the log does not tell you how long analysis, implementation, or work outside the interface took. I would not pretend that four entries equal one hour of labor.
Now compare the categories with performance. If onboarding showed sustained work but the current month shows little beyond automated entries, while CPA and qualified pipeline have stalled, the agency has an explanation to provide. If the log is quiet and the account is meeting its CPA target while qualified pipeline grows, you may be looking at restraint rather than abandonment. Ask what the team monitored, what it chose not to change, and whether consequential conversion or offline work happened elsewhere.
I used to give clients a version of the mature account speech when I was juggling too many accounts. It has a real technical point: reckless changes can disrupt a bidding strategy. It also makes an excellent hiding place for a team that checks pacing once a week and calls the absence of fires a strategy. Markets move. Search terms drift. Landing pages change. “We left it alone” is a decision only if someone can show why leaving it alone served the business.
This is where the controls matter. An empty log alone cannot prove nobody worked. A busy log alone cannot prove anybody helped. The combination to challenge is an extended performance plateau, few defensible interventions, and no credible account of what the agency did instead.
Book the review before the next billing cycle. Bring the exported log, your three comparison windows, and the business baseline. Keep the conversation concrete:
A capable media buyer should be able to walk you through the decisions, acknowledge a lull, or show you meaningful work the Google Ads history missed. If the answer is a tour of Optimization Score and a claim that auto-applied changes count as attentive human management, you have learned something useful about the retainer.
This is the operating-model question behind the audit. Ad auctions keep running between status calls. groas replaces periodic manual execution with specialized AI models that work continuously on bids, targeting, budgets, and creative, while a human strategist sets direction and guardrails. That is a better fit than paying for sporadic oversight dressed up as constant optimization. The change log gives you a way to test the distinction rather than take either pitch on faith.
Run the protocol every quarter, whether an agency, an in-house specialist, or autonomous software manages the account. If the record shows deliberate interventions tied to improving qualified pipeline, keep the operator and give the work room to run. If performance is stalled and the log offers only silence and automated filler, stop renewing the retainer on the strength of a deck. Ask for an answer before the next invoice; if it does not hold up, move the spend to an operating model that will act on what the account needs.
How can I check what my agency actually changed in my Google Ads account?
Open Campaigns and select Change History in the Google Ads desktop interface, which keeps a rolling two-year record. You need standard or administrative access; if the agency controls the account and will not grant it, resolve that first, because you cannot audit an account from a PDF the agency chose to send.
What time periods should I compare when reviewing the Google Ads change history?
Set the test period to 30 completed days ending two days ago, then pull two comparison windows: the 30 days immediately before it and the first 30 days of the agency's engagement. Use the same account and classification rules for all three, and write down the business baseline for each window before scoring anything. Excluding the most recent two days gives conversions time to appear.
Why shouldn't I just count the number of changes in the log?
A raw count rewards busy-looking activity: one bulk action can generate many entries, automated changes can fill a month, and a renamed ad group is not strategy. Instead, label each entry as automated activity, cosmetic or clerical activity, or a high-impact intervention, and apply one consistent test for ambiguity: which commercial variable could this edit change?
Do auto-applied recommendations in Google Ads count as real optimization work?
No. An agency does not get to call auto-applied recommendations hands-on optimization merely because it enabled them, and unmonitored rules can make consequential changes on their own. Note who configured and monitors the automation, but the useful question is who reviewed those changes against your economics.
What should I look for when scoring high-impact interventions in the change log?
Look in four places: queries and exclusions such as negative keywords, bidding and value-signal changes like target CPA or target ROAS adjustments, budget shifts between campaigns, and new ad versions or replaced headlines and images. An edit counts only when it connects to a specific problem the agency was trying to address, and these buckets are not a quota.
How do I price the interventions my agency made?
Count the high-impact interventions in your test window and divide your retainer by that count; a $4,000 retainer with four interventions is $1,000 per logged intervention. Treat this as a prompt for a hard conversation, not an hourly rate, because the log does not show how long analysis or work outside the interface took.
Does a quiet change log mean my agency isn't doing any work?
Not on its own. An empty log cannot prove nobody worked, and a busy log cannot prove anybody helped. What should concern you is the combination of an extended performance plateau, few defensible interventions, and no credible account of work the log missed, such as conversion tracking or investigations that led to a deliberate decision not to edit.
When should I confront my agency about the change log findings?
Book the review before the next billing cycle and bring the exported log, your three comparison windows, and the business baseline. Ask which changes were meant to address the CPA or pipeline plateau, why the mix of interventions changed after onboarding, what work happened outside the log, and what test comes next with which metric.
