DataForSEO misses half the AI Overviews I tested. For 2 out of 4 queries, an empty block where the other API returns the full answer, sources included.
So the question lands cash on the table: can I actually trust a single SERP capture to track my AI Overviews? Because if the tool feeding my SEO reporting drops half the signal without telling me, everything I build on top of it is quietly wrong, and I have no way to know.
I sent the same 7 queries in parallel to both APIs, google.fr, desktop and mobile, depth 3. Not the marketing docs from either company. Raw data, straight out of the pipe.
I Sent 7 Queries to 2 APIs
Batch specs, so nobody has to guess at the setup later:
7 queries, google.fr
Desktop + mobile
Depth 3
Google Maps pack included
I picked queries that reliably trigger an AI Overview on google.fr right now, mixed commercial and informational intent, nothing exotic. The point was never to stress test edge cases. The point was to see what 2 APIs report back on the exact same search, at the exact same moment.
Cost, for the record: SemScraper ran 0.035€ for the whole batch, 0.005€ per query. DataForSEO ran about 0.014$, roughly 0.002$ per query. This is not a budget test, it's a signal test. I got into SERP APIs in the first place after fighting CAPTCHA blocks trying to check rankings by hand, so paying a few cents to skip that fight entirely was never the hard part. This is not a boss fight. It's an API key and a coffee break (minus the coffee break).
Same Rankings, Then It Splits
On plain organic rankings, both tools agree almost perfectly. Same 9 domains near the top, same order, across all 7 queries. That part of SERP scraping has been commoditized for years. Nobody wins or loses on domain positions anymore.
Then I checked the AI Overview block. SemScraper returned a complete Overview on 4 out of 4 queries where Google actually shows one. DataForSEO returned an empty block on 2 out of those same 4.
On the 2 queries where both tools did catch the Overview, the cited sources line up closely between the 2 responses. Same domains, same order, roughly the same snippet text. That detail matters. If DataForSEO's parser was reading the block wrong, I'd expect garbled or mismatched sources on the ones it does catch. It doesn't. The block is either there or it isn't.
4 against 2 looks like a closed case. But is a single snapshot even the right way to judge whether an API is good at this?
Why One API Misses What the Other Catches
The gap here is not about which company writes better scraping code. It's about timing.
Google generates AI Overviews asynchronously. The block isn't always ready the instant a page loads. A crawler that renders once, grabs the DOM, and moves on can hit that exact page in the half second before the Overview populates. The AIO exists on that query. The bot just showed up too early, like walking into a cutscene half a beat before it loads.
I think that's the actual mechanism at play here, though I could be wrong on the exact retry logic each vendor runs under the hood. Neither company publishes that part. That timing gap explains why a single-shot API is structurally worse at this specific job, no matter how solid its infrastructure is otherwise. If a scraper fires once and walks away, it inherits Google's own render delay as a coin flip on every AIO query, and that coin flip compounds across a full keyword list the way any hidden failure rate compounds, quietly, until someone actually checks the raw output instead of trusting the dashboard summary.
A tool built to wait, retry, or poll for the block before giving up has a structural advantage here that has nothing to do with data quality and everything to do with patience baked into the request loop. That's the part vendor comparison charts never show you, because it doesn't show up until you run the same query enough times to catch the block missing.
What Bright Data Admits About the Same Problem
Bright Data wasn't part of my live test. I didn't run queries through their API myself, this section comes from their own documentation and public pricing, not a side by side run. Worth stating plainly before the numbers.
3 things stand out. I already wrote my full breakdown of Bright Data's pricing and quirks after evaluating them for a different project, and this section leans on that same research plus their current docs.
Price: Bright Data's SERP API runs 1.50 to 3 dollars per 1000 requests depending on the mode. DataForSEO's Standard tier runs about 0.55 dollars per 1000. Bright Data costs 3 to 5 times more for the same job.
Positioning: Bright Data sells this enterprise style, sales calls and contracts, not a self serve dashboard you sign up for on a random Tuesday night.
The admission: their own docs describe the AI Overview parameter as something that increases the likelihood of capturing the block, with a typical rate they list at 15 to 20 percent or slightly higher. Not a guarantee, just a likelihood, on paper, from the vendor itself.
That's the part worth sitting with. The most expensive option in this comparison, the one built for enterprise contracts, states in its own documentation that AI Overview capture is a probability game, not a solved problem.
Enterprise pricing doesn't buy you certainty. It buys you a nicer probability.
If even the priciest vendor on the list admits partial capture on this exact point, is price the thing that should decide this, or is everyone just guessing at slightly different odds?
The Bigger Problem Isn't Which API You Pick
Before February 2026, whichever API caught more AI Overviews was basically the whole story. After February 2026, that stopped being enough on its own.
Since then, Google has been serving deliberately falsified SERPs to some tracking bots. Pages stuffed artificially with video content, results that don't match what a real user sees on the same search. The reporting on this comes largely from Monitorank, the company behind SemScraper, the tool that won my test above. Worth flagging that plainly (they have a stake in this story looking a certain way). Some tools patched their detection within days. Others kept feeding corrupted numbers into client dashboards for weeks before anyone noticed the pattern.
Random unrelated thing: my downstairs neighbor started renovation work this week, drilling through the exact hours I do my writing. No connection to shadow SERPs, just background noise while you read this.
Then in May, a second episode. The gl parameter, the one that tells an API which country's Google to query, started returning results that didn't match the country requested. Less reporting on this one came from a party with a direct interest in the outcome, that part is documented more independently.
2 separate incidents, 6 months apart, same underlying theme. Whichever API wins on AI Overview capture rate this month, the ground both of them stand on keeps shifting without warning.
What I'm Actually Using Now
SemScraper wins this specific test. 4 AI Overviews captured out of 4, DataForSEO caught 2. On the exact question of tracking AI Overviews reliably, that's the tool doing the job right now.
DataForSEO isn't going anywhere for me though. Their Labs data, search volumes, keyword suggestions, SERP intersections (that's a different product entirely), and this test never touched any of it. Nothing here says switch everything.
Bright Data stays a question mark. Everything I wrote about them above comes from their public docs and pricing pages, not from a side by side run on my own queries. Treat that section as secondhand, not verdict. Calling any of this a final ranking would be generous anyway, it's closer to comparing loot drop rates before you've even picked a class.
That verdict holds for what it tested. Nothing more.
Because the real subject already stopped being which API to pick. Since February, Google has been serving deliberately falsified SERPs to tracking bots. In May, the geo targeting parameter went quiet without telling anyone.
What's left now is how long a piece of SERP data stays SERP data before Google decides to poison it under your feet.
Sources
- SERP API documentation, Bright Data
- Bright Data SERP API, pricing and billing
- Shadow SERPs: Google poisons tracking tool data since February 2026, Atom-Business (reporting largely sourced from Monitorank, the company behind SemScraper)
- Suivi de position: pourquoi vos données Semrush ne reflètent plus la SERP française, Eric Garletti
This post may contain affiliate links. If you click them, I might earn a small commission (costs you nothing, and helps me keep shipping quality articles every day for your reading pleasure).
Top comments (0)