An astute reader emailed me (Preston, you're a genius) to suggest that rather than attempt to collect PPC stats on our own, why not tap into the vast resource of the Internet marketing community - surely, together, we will be far stronger than we would be as just one firm.
Here's the goal - to find a formula whereby we can measure the efficacy and accuracy of keyword research sources like Overture, KeywordDiscovery, Hitwise's Keyword Intelligence, Google's Adwords Estimator, MSN's Adcenter, and others. There's just a few simple steps to follow:
- Gather a large list of hundreds, hopefully thousands of keyword terms and their actual search numbers at Google - this can be done by getting data from advertisers who were always listed on the 1st page of results for the term through AdWords (using the impression numbers for "exact match").
- Gather a list of the estimated numbers reported by all of the major keyword research services for those same terms and phrases.
- Distill the data, tossing out those phrases that might cause issue or those reports that wouldn't have a huge impact and keep only the best, most useful ones.
- Compare the two sources (Adwords Impressions and estimator tools) and find out fascinating information like - which tool is most accurate? Is there a formula we could create that would predict (based on all the estimation sources) within 10%, 20% or 30% of actual traffic?
To my mind, this formula would be priceless to search marketers. Once you can accurately predict search traffic (even if it's off by 20-30%), you've got a system that doesn't require a massive spend at Google to predict actual searches. Even if the data isn't conclusive, the results will be incredibly enlightening.
We have a few clients (and some botched test data) that we can provide for the project, but we really need your help to make it possible. My promise is that we'll collect the data, report it all honestly (but anonymously - not revealing any of your keywords or those of clients) and build a tool that will use this formula. We'll also provide the exact formula publicly, so that anyone can use the same information to build their own tools, use it in their internal calculations, etc.
How can you help?
- If you are willing to participate, please email [email protected]. We'll figure out all the necessary details, whittle down what's required, and request screenshots to help make the data collection happen.
- If you are a PPC expert and can help us to analyze the data and issue the requirements, please email Rebecca as well; we'll need some help from the best :)
- If you're a blogger, you could ask your readers to do likewise (you don't need to link here, you can just ask them to email Rebecca; I don't want it to look like I'm begging for link love)
- If you participate at forums like DigitalPoint, Sitepoint, WebProWorld, SearchEngineWatch, HighRankings, etc. you could post about this and ask for volunteers (again, no need to link).
- If you work for Google, you could just make the AdWords estimator accurate and save all the trouble (we'll even send you a lovely present).
What do you say? I think that together, we could make something that will be truly valuable for years to come (and, as the data becomes less accurate, we can refresh it).
Am I off my rocker here, or is this an idea that has some merit? Any suggestions, additional ideas, important points to consider?
p.s. Rebecca - I'm sorry to flood your inbox like this 
While a noble idea, it seems like a lot of work with no real payout, other than maybe some linkbait ;)
I mean is anyone really going to change the way they are performing SEO because some tool said they can expect 50K searches a month at the top and in reality there are only 29K? In the end, completely whitehat websites are trageting their keywords because that is what they are about and there are at least a few people searching for those terms. So in the end, they are going to target the terms they are relevant to regardless of the traffic.
Even if someone did need to know a more accurate figure for a set of terms, they can run their own test. I have actually used this method to determine if a niche was worthy of building a site for.
Sometimes we as SEO over analyze things and it is not cost effective.
I think this data is actually more useful to arbitragers and blackhats than to true whitehat websites.
You're lucky I love my job and like feeling important *grumble grumble* ;)
Rand:
The "noise" issues are not going to be a problem, because most real users only go through the top 3-5 listings in SERP anyway.
The real issue is that estimator tools take a different approach to broad/phrase/exact matching.
A list of terms including widgets, blue widgets and green eyed widgets will give you 3 numbers that vary in accuracy. The count for widgets will be broad while the count for green eyed widgets is inherently narrow (if not really a phrase or exact match).
Pay really careful attention how the terms are setup in adwords before you try to compare the data. You may be able to run the search for widgets, "blue widgets" and "green eyed widgets" on some tools to isolate the impact of this problem.
I must agree with the other naysayers. We have tried to do this many times with the data we get from PPC clients. First, one must have an unlimited budget to have their ads show up everytime the search is conducted (meaning the big dogs would have to share data - ha). Then, all the other starts must align like the others point out (the phrases must be exact match, geotargeting turned off, people don't view the second page of the results, and so on). Every time we have done this, there has been no definite pattern to apply to one specific keyword tool in any way.
That being said, I do think PPC data is good at showing the volume of the searches that an exact match phrase has compared to another phrase. While the numbers might not be accurate you can determine that you should get X% more volume from this term vs. that one organically based on PPC data.
However, if you guys pull this off and get some great formula for us to use, consider it a guaranteed link from me for your great work!
Hi Rand, our agency actually has quite a few PPC clients covering a wide range of search terms and volumes, and a few months ago I’ve already trolled though all the PPC data a while ago with the same object in mind.
However, the problems of noise, broad match, content ads, ad times, seasonal changes, location, and other factors described in other comments leaves behind a quite unreliable result.
I agree to accurately predict search volumes would be priceless, however even when I had a huge volume of PPC data at my disposal, there was too many external factors that prevented me from making an accurate prediction.
I don't know how useful it would be, but our database of keyword difficulty reports contains over 20,000 terms with the number of searches each month according to overture.
It's a very good idea, in theory. A WikiStats of sorts. But are you planning to show your customers the real real data? The online advertising market already prides itself on better than average statistics, but if this kind of tool was implemmented it would mean people actually had to be accountable for every penny spent... scary thought... ;)
One problem with adwords... your ad shows on multiple pages of the SERPs. I wonder if stats from inventory.overture.com and other sources are tainted by this problem?
When I want to know something about "relative" search volume, I use google trends.
Seems a good idea. You would also need to collect geotargeting data.
To provide more details about what EGOL said, if Google shows my ad in one of the blue positions at the top, it will show on every search results page. (The sidebar ads don't repeat from page 1 to 2, but the top ads do.) That means I'm counting two searches if users went to the second page. I've run into this issue before when trying to get a gauge on search volume for keywords :(
Is there any workaround for this? Could we get data that would allow us to know how many impressions on average come from multiple page views to help cancel out that issue?
Rand, we've done this before and I actually benefit from these numbers. The way to do this is (as was mentioned ... "no geotargeting, unlimited budget, etc) is also to bid just enough to be somewhere on the bottom of the PPC results. #7 to 9 works for me.
When I would compare this info I always found that data from tools, such as WordTracker, etc, is off.
I remember I blogged about this saying that before people even start their SEO campaigns they should run PPC campaign for all the variations of keywords and then decide which keywords to target based on their traffic AND conversion (browsing behavior).
Few other things to consider is not doing this during any seasonal holidays as well as avoiding keywords with high possibility of click fraud. For some of our keywords I'd sometimes see October show 150,000 impressions for a certain keyword and the next month it would be 60,000. .... Content match turned off.
I think this is a 2 months minimum project. Unless you want to start dividing monthly numbers into weeks.
This data is very noisly... other variables... some searchers have their Google preferences set to ten search results per SERP... others might use one hundred. How are scrapers and bots counted... lots of them to muck up the numbers and the activity of these is not uniform across the web.
My approach is to put up some test ads pointing to a good landing page and see what happens. I do this before running adwords and before sinking a lot of resources into building labor intensive content.
Rand - I have often thought about ways to try to collaborate on keyword data. Most companies feel this is highly proprietary data, and for good reason. Most SEO firms are probably not able to divulge this data due to their client contracts.
I think this project has great potential, and i may participate. Just know that everyone is holding back sensitive keyword data.
CVOS - completley understood. We'll have to rely on the kindness and trust of everyone in the project. I think we can find at least a few people who are willing to take a chance for the benefit of the community as a whole. Hope to see you involved, too :)
I think the data can be very useful. I work primarily in the computer hardware industry and when I target a keyword I typically take Yahoo's inventory tool and multiply that by a certain # to estimate Google's volume - within my industry this is pretty accurate (or at least seems to be).
Recently, however, I did a project for a client outside of the industry and assumed the same multiple - I was way off. Perhaps my industry is an anomoly and by seeing how the aggregate data stacks up I could make better decisions on keywords to target for clients in different categories.
Seems there is a more accurate means of determing search volume form keywords. Yet, it takes collaboration just as the proposals here take that into account.
One the most accurate stats on a web site is the "key word" that brought search engine traffic to your web site.
Now if someone could figure out the scheme to attract these results into a single database the game would be so to speak, ON!
Wow, I don't know about you guys, but for me, I always have a hard time trying to give accurate predictions for my clients. How many visits will they receive from pay-per-click, what king of budget should they consider to be valuable, etc. So, without a doubt, this is the kind of tool that would make my life so much easier. I'm sure in! But 80% of my data is coming from french Canada, so I hope we will get several players in Canada, USA and Europe to make this thing valuable to everyone.
Salut à toi chez Ressac!
I think that the fact French Quebeckers is 30x of a smaller market than English in North America makes it very difficult for us to give great predictions considering we get very low keywords impressions for small niches and also most of the keywords tools are simply ENGLISHLINGLY rigged...
On our personal side at NVI, we've seen alot of Quebeckers who search Brands related keywords (a great example is Multi-Prêts, a mortgage firm, which has so much more searches for the brand itself than all "industry related keywords" put altogether).
My call for this is that verticals data for keywords will greatly vary, depending on the "demographic/behaviors" of the databases queried for each keyword tool.
I'm more than willing to be a part of this though ;)