For two weeks we recorded no Overview on 193 of 201 searches
Between 22 August and 7 September 2026 our sampler put 36 questions about hiring software to Google, from a US location, through DataForSEO's search results API. It logged 201 searches. 193 of them, 96.0% (95% interval 92.3% to 98.0%), came back with no AI Overview text at all. Neither the tracked brand nor any of its competitors was named once.
Read at face value, that says Google rarely shows an Overview for these questions, so there is nothing to win there. It was wrong. Google was showing one on almost every search. We weren't loading it.
Google loads many Overviews after the page, and the default fetch skips them
DataForSEO's documentation describes two kinds of AI Overview. A synchronous one is already in the HTML when the results page loads. An asynchronous one appears a moment later, when it needs more time to generate. The API's load_async_ai_overview parameter defaults to false, and with it off you get only the Overviews Google had cached. Turning it on costs an extra $0.002 per request, refunded when no asynchronous Overview turns up.
This isn't one vendor's quirk. SerpApi's documentation says Google may lazy-load the Overview for less common or more complex queries, and that the follow-up token needed to fetch it expires within a minute. DataForSEO's own December 2025 update says its datasets had previously captured only cached Overviews, and that Google frequently serves them asynchronously.
Our sampler had the parameter off. We turned it on in a fix on 8 September, and left that changeover day out of both counts below.
On the same 36 questions, 3 of 487 searches came back empty
From 9 September to 5 October the same 36 questions, same location, ran 487 more times. Three came back with no Overview: 0.6% (interval 0.2% to 1.8%). Across a wider set of 485 questions in 14 categories, asked between 21 September and 5 October, it was 38 of 1,125, or 3.4%.
The answers themselves changed what the brand could see. Before the fix, no competitor appeared in any Google answer. After it, at least one tracked competitor was named in 41 of the 484 Overviews, 8.5%. A team reading the old numbers would have concluded that nobody in the category was visible on Google, when the competitors simply hadn't been loaded.
A missing answer scored as a miss looks exactly like losing
Most visibility scores divide answers that name you by searches run. If a search with no Overview counts as a search where you weren't named, every failed load pulls your score down. If the failure rate shifts from week to week, as vendors and Google both change, your trend line moves for reasons that have nothing to do with you.
The same trap applies outside Google. An engine sometimes answers with a question of its own: which category do you mean? Since 28 September we have recorded 52 of these in 4,139 answers across four engines, 38 of them from ChatGPT. At least 39 of the 52 came from questions we had written without naming the category, such as "what companies are well known in this space", so the fault was largely ours. Counted as misses, they would read as the engine ignoring the brand.
Proofsource now treats both as no answer. A search where Google showed no Overview, or where the engine asked back, is left out of visibility and share of voice. Since 2 October, an Overview that Google showed but the vendor failed to load is retried and, if it never loads, recorded as a failed search, not an empty one.
Three questions to ask any tracker before you trust its Google numbers
These are short questions, and a vendor that measures carefully will answer them in a sentence each.
- Do you load asynchronous AI Overviews, and since when? Any history from before that date understates Google.
- When no Overview appears, is that search left out of my score or counted as a miss?
- Can I see the rate of empty searches per week? If it jumps, the score is about to move for a reason that isn't me.
Where this could be wrong
The before and after comparison covers one set of 36 questions in one category from one location. The 96.0% describes our setup with the parameter off, not how often Google uses asynchronous loading in general.
The two periods are a few weeks apart. Google could have changed how often it showed Overviews in that time, though a fall from 96.0% to 0.6% across the day a fix shipped is hard to explain any other way.
We can't tell how other trackers handle this. Some may have loaded asynchronous Overviews from the start. The point of the questions above is to find out.
Some of the 3.4% may be Google genuinely not showing an Overview, and some may be load failures from before 2 October that we couldn't yet tell apart.
