Field guide
Pulling Google text ads and Shopping results in bulk: what four SERP APIs returned for 25 keywords
25 US commercial keywords, four SERP API routes, two fetches each from New York on one day: how often a text ad came back at all, how many advertisers each route saw, how much changed between fetches minutes apart, and which route returned the sponsored Shopping block.
Written for people asking: Any tools that pull US paid results (text ads and Shopping) in bulk?Are SerpApi, DataForSEO, Semrush or SpyFu accurate for competitors' Google ads?Is Google Ad Preview a fair picture of the live auction?How do I see which competitors bid on a keyword in Google Ads?
A SERP API gives you a sample of the paid half of Google's results page, not a census, so plan on several routes or several fetches per keyword. On 8 November 2026, 25 US commercial keywords went through four routes twice each from New York, desktop, 200 fetches in all. Text ads came back on only 28 of the 199 fetches that answered, and 11 keywords, including personal loan, mortgage refinance rates and emergency plumber, returned no text ad on any fetch from any route. Across everything, 38 keyword-advertiser pairs showed up on 14 keywords; 30 of them were seen by only one route. SerpApi saw the most, 20 of 38 over two fetches (16 on one), DataForSEO 18, Cloro 5 and Litescrape's ads route 4. Fetching the same keyword through the same route minutes later changed the advertiser set almost completely (overlap 0.21 on SerpApi, 0.15 on DataForSEO). For Shopping, only Litescrape's ads route returned the sponsored product block, on 5 of 7 product keywords and on both fetches; the other three returned only Google's unpaid product grids. Settled prices per fetch were $0.015 SerpApi, $0.002 DataForSEO and Cloro, and $0.00015 Litescrape; all four ran through Treg with one key, about $0.96 for the 200 fetches.
The job, in one line
A PPC buyer has about 150 keywords before a launch and wants to know who is bidding on them in the US: the text ads above and below the organic results and the Shopping products. The question went up in about ten PPC and SEO communities in an hour, naming SerpApi, DataForSEO, Semrush and SpyFu and asking whether any of them is accurate. Semrush and SpyFu sell their own historical ad databases and were not reachable as routes here. What can be measured is the raw material those tools and your own scripts start from: what a SERP API returns for the paid half of the page, per fetch.
What came back
Twenty-five US commercial keywords across software, insurance and finance, legal, home services, travel and retail products (fixture), each fetched twice through four routes on 8 November 2026 (UTC), from New York on desktop where the route takes a location. The second fetch followed the first by a few minutes.
| Route | Fetches answered | Fetches with a text ad | Text ads (top / bottom / other) | Keywords with an ad | Advertisers seen, of 38 | Sponsored Shopping fetches | Price per fetch |
|---|---|---|---|---|---|---|---|
| SerpApi (Google engine) | 50 of 50 | 12 | 30 (20 / 10 / 0) | 9 | 20 | 0 | $0.015 |
| DataForSEO (organic live advanced) | 50 of 50 | 10 | 23 (7 / 16 / 0) | 8 | 18 | 0 | $0.002 |
| Cloro (Google SERP) | 50 of 50 | 4 | 5 (0 / 2 / 3 “middle”) | 4 | 5 | 0 | $0.002 |
| Litescrape (Google ads) | 49 of 50 | 2 | 4 (0 / 4 / 0) | 2 | 4 | 10 | $0.00015 |
“Advertisers seen” counts keyword-advertiser pairs, by the advertiser’s own domain, across both fetches; 38 is the union over every route and fetch. One Litescrape fetch failed with service_unavailable and was not billed. Cloro takes a country but no city, so its fetches are US-wide, not New York. Prices are what the Treg ledger settled.
Most fetches carry no text ad
Of 199 fetches that answered, 28 had a text ad. Eleven keywords returned none on any of their eight fetches: personal loan, business credit card, mortgage refinance rates, emergency plumber, roof replacement cost, pest control service, standing desk, air purifier, mattress, office chair and espresso machine. These are terms people buy ads on; why the fetched pages came back without them (location, the fetch looking automated to Google, or the auction itself) is not established by this run. Whatever the cause, an empty ads array from one fetch does not mean nobody bids on the keyword.
The same route, minutes later
Between the two fetches the paid block mostly turned over:
- Presence flipped. On 6 of 25 keywords for SerpApi and for DataForSEO, one fetch had ads and the other had none. “Email marketing platform” returned six SerpApi ads, then two; DataForSEO returned none, then six.
- Different advertisers. Where either fetch had ads, the share of advertisers common to both averaged 0.21 on SerpApi and 0.15 on DataForSEO (Jaccard), and 0 on Cloro and Litescrape.
- A second fetch adds more than a second route. SerpApi saw 16 pairs on the first fetch and 20 over both; DataForSEO 7 on the first and 18 over both. SerpApi and DataForSEO together over two fetches saw 32 of the 38.
Thirty of the 38 pairs were seen by one route only, seven by two and one by three. That is the practical answer to “is any of them accurate”: each returns a real page, and each page is one draw.
Shopping: the sponsored block versus the free grid
Google shows two kinds of product boxes: the sponsored product carousel, and unpaid grids of popular products. All four routes returned unpaid grids, on 14 to 25 of their 50 fetches (immersive_products on SerpApi and Litescrape, popular_products and product_considerations on DataForSEO, shoppingCards labelled “More products” on Cloro). Only Litescrape’s ads route returned a shopping_results block, which is the SerpApi-schema name for the sponsored carousel: 8 to 10 products with merchant, price and product id for standing desk, wireless earbuds, air purifier, mattress and espresso machine, on both fetches. Running shoes and office chair had none. The block carries no explicit “sponsored” label in the response, so the reading rests on the schema name. SerpApi’s own route, with the same schema, returned no shopping_results for any of the 50 fetches.
Top versus bottom
SerpApi and Litescrape label each ad with block_position; DataForSEO gives a rank on the page, read here as top when it sits above the first organic result. SerpApi’s ads were two-thirds top (20 of 30) and DataForSEO’s two-thirds bottom (16 of 23). Cloro labels three of its five ads “middle”. If your report splits top from bottom, take that split from one route, because the routes disagree on it.
What a 150-keyword read costs
| Plan | Fetches | SerpApi | DataForSEO | SerpApi + DataForSEO |
|---|---|---|---|---|
| One fetch per keyword | 150 | $2.25 | $0.30 | $2.55 |
| Three fetches per keyword | 450 | $6.75 | $0.90 | $7.65 |
Add Litescrape’s ads route for the sponsored Shopping block at $0.00015 a fetch, $0.07 for 450. At these prices the second and third fetch cost less than the analyst time spent arguing over one.
What this run cannot tell you
One day, one city, desktop only, 25 keywords and two fetches each, so the rates above are a pattern rather than a benchmark. There is no ground truth for the live auction in this run: Google’s Ad Preview tool, Semrush and SpyFu were not measured, and a human search was not taken alongside. Advertiser domains are read from each ad’s displayed link where the link itself is a Google click URL. Ad copy, extensions and landing pages are in the raw responses but not scored. For the advertiser side of the question (every ad a known competitor has run), see the Google platform page, which reads the Ads Transparency Center by domain.
How this was measured
- Fixture: paid-serp-keywords-v1.json.
scripts/run-paid-serp-check.pymakes the 200 calls through the treg CLI, one worker per route, and stores one raw response per call.scripts/score-paid-serp-check.pyreads them with no provider call and writes summary.json and keywords.csv: per keyword, fetch and route, top and bottom text ads, advertiser domains, sponsored Shopping products and whether an unpaid product grid was present.- The whole run settled about $0.96 on the Treg ledger: $0.75 SerpApi, $0.10 DataForSEO, $0.10 Cloro and under a cent Litescrape.
Next step
Read the paid block for one keyword through two routes and compare:
treg auth login
treg call serpapi.google.serp.organic --query engine=google --query 'q=email marketing platform' --query 'location=New York, New York, United States' --query gl=us --query hl=en --query no_cache=true > s.json
treg call litescrape.google.serp.ads --query 'q=standing desk' --query 'location=New York, New York, United States' > l.json
python3 -c 'import json; s=json.load(open("s.json")); l=json.load(open("l.json")); print("text ads:", [(a["block_position"], a.get("displayed_link")) for a in s.get("ads", [])]); print("shopping:", [(p["source"], p["price"]) for p in l.get("shopping_results") or []])'
Or run your own list: put your keywords in the fixture format, then scripts/run-paid-serp-check.py <fixture> evidence/paid-serp-checks/$(date +%F) 3 and scripts/score-paid-serp-check.py <fixture> evidence/paid-serp-checks/$(date +%F). Download the runner · the scorer.
Questions this page answers
Which API pulls Google text ads and Shopping ads in bulk?
No single route returned both well on 8 November 2026. For text ads, SerpApi and DataForSEO returned the most: over two fetches of 25 US keywords from New York, SerpApi saw 20 and DataForSEO 18 of the 38 keyword-advertiser pairs any route saw, 32 between them. For the sponsored Shopping block, only Litescrape's ads route returned it, on 5 of 7 product keywords on both fetches; SerpApi, DataForSEO and Cloro returned only Google's unpaid product grids for the same keywords.
How accurate are SERP APIs for competitors' paid search ads?
Each fetch is one sample of an auction that changes between fetches. The same route fetching the same keyword minutes apart kept on average 21% of its advertiser set (SerpApi) or 15% (DataForSEO), and 30 of the 38 advertisers seen across four routes were seen by only one. Eleven of 25 commercial keywords returned no text ad at all in 200 fetches. Count a competitor as bidding after several fetches, not one.
Is Google's Ad Preview tool a fair picture of the live auction?
This run did not test Ad Preview, so there is no measured answer here. What it shows is that any single look at the results page, by a tool or an API, is a sample: the advertiser set for the same keyword changed almost completely between two fetches a few minutes apart. Repeat the look before you size an auction from it.
Relationship: the people who run this site also work on Treg, one of the routes compared on this page. Every Treg figure here is a call we paid for, with the raw response downloadable, and the same fixture is run against the alternatives where their terms allow.
Spotted an error or a changed price? Email corrections@adapiguide.com with the page URL and a source. Corrections are logged on the corrections page.