“Custom” Datafeed Descriptions

Two things before we get started:

1. I love datafeeds. How great is it that merchants give you a list of hundreds or thousands of products with titles, descriptions, categories, URLS, prices, etc. to use on your site? That is instant content I don’t have to create.

2. I am not a programmer at all. Anytime I try to simplify anything by mashing together some scripts or Excel files or whatever it usually takes me much longer than if I just did it manually.

With that out of the way let me get into one of the problems with datafeeds. That is the problem of creating custom content for the products in a datafeed. If you want your datafeed site to be successful you will usually need to get some sort of custom content in there and not just use the datafeed straight from the merchant. Chances are the content in their datafeed is already on their site and on dozens of other datafeed based sites. So custom content in your datafeed will give you a leg up on the competition.

If you are working with a datafeed of a couple dozen or even a hundred products it really doesn’t take that long to custom write descriptions for each product. If you were to sit down and really crank them out you could probably do 15-20 of them in an hour or even more depending on the length of the description. But what about when you have thousands of products? Or even hundreds of thousands?

I have sent datafeeds to some outsourced workers in the past and have gotten some pretty decent custom product descriptions written for about 20 cents each for 50-100 word descriptions. But it takes time for them to do the writing and 2000 descriptions will cost you $400 which is a pretty big investment. In the long run it can be worth it but then you run into a problem a month down the road when the merchant adds some new products or changes the price or something and you have to go in and update your feed manually. I am sure there are some pretty easy ways to automate this type of thing but like I said, I am pretty stupid when it comes to stuff like that and I usually end up screwing it up.

Another option is to just use a spinner but it can be pretty hard to find a good spinner and even the good ones aren’t really that good.

Well, a few days ago an article was posted on SEERInteractive.com by Brett Snyder about how he wrote 500,000 unique GoogleBase descriptions in 2 hours. He was seeing some pretty bad descriptions for some products and wanted to rewrite them so he came up with a pretty awesome way to do that using Excel.

When working with datafeeds Excel is your best friend. I am always searching, replacing, sorting, and so on with Excel but when I see stuff like this I realize more and more how powerful Excel really can be. Basically what Brett was doing was using Excel to randomly pull in sentences and keywords and build “custom” product descriptions.

What I really liked about this method is you could write 20-30 sentences or so, mash them all up, and have some descriptions that will be different than the standard descriptions the merchants give with the datafeeds. And the best part is he included a link to the Excel template he uses so pretty much you just need to replace the content he was using with your own. I won’t go into too much detail about how that template works here. You can read his article to get that info.

Here is how I used this to write over 1400 product descriptions for a feed I was working on:

1. I broke the product name into some keywords. The feed I was working with was for NFL sweatshirts. Kinda weird because the example Brett uses in his article is NFL jerseys. Anyway, here is what the regular product names look like:

• Reebok Seattle Seahawks Royal Blue Vintage Team Logo Pullover Hoodie Sweatshirt
• Seattle Seahawks Ladies Navy Blue Football Classic III Full Zip Hoodie Sweatshirt
• Seattle Seahawks Ladies Steel Blue Game Day Heroes II Pullover Hoodie Sweatshirt
• Seattle Seahawks Ladies Steel Blue Overtime Queen II Split Neck Heathered Sweatshirt
• Seattle Seahawks Charcoal 1st & Goal IV Hoodie Sweatshirt
• Seattle Seahawks Navy Blue Overtime Glory Full Zip Hoodie Sweatshirt
• Seattle Seahawks Navy Blue Passing Game II Pullover Hoodie Sweatshirt
• Seattle Seahawks Navy Blue Classic Heavyweight II Full Zip Hoodie Sweatshirt
• Seattle Seahawks Black Touchback IV Full Zip Hoodie Sweatshirt
• Seattle Seahawks Navy Blue Classic Heavyweight III Pullover Hoodie Sweatshirt
• Seattle Seahawks Royal Blue Critical Victory V Pullover Hoodie Sweatshirt
• Reebok Seattle Seahawks Youth Navy Blue Active Pullover Hoodie Sweatshirt
• And so on. There were 1462 total products in this feed

I cut and pasted all the product names into a new file so I could work with it more easily than in the full datafeed file.

Looking at those product names I needed to figure what keywords I could pull out, which ones were important, and how they could all be re-combined to make sense in the final descriptions. Obviously I could pull the team name. The feed had a column for the team name already so I just deleted all the team names out of the product name field.

Then I looked at what type of sweatshirt they were and split them into “hoodie” and “crewneck” since pretty much all of them fell into one of those two categories. To do this I set up a filter in the product name column to look for the word “hoodie”. When they were all filtered I created a new column and just put the word “hoodie” in it then unfiltered, sorted by the “hoodie” column and named everything else “crewneck”. Then I removed “hoodie” from the product name column.

I repeated this process a couple more times for stuff like full zip vs pullover, the manufacturer, if it is for men, ladies, kids, etc. My goal was to get pretty much everything out of the product title except the actual product name. So when I was finished my product name column looked like this:

• Royal Blue Vintage Team Logo 
• Navy Blue Football Classic III 
• Steel Blue Game Day Heroes II 
• Steel Blue Overtime Queen II Split Neck Heathered
• Charcoal 1st & Goal IV
• Navy Blue Overtime Glory 
• Navy Blue Passing Game II 
• Navy Blue Classic Heavyweight II 
• Black Touchback IV 
• Navy Blue Classic Heavyweight III 
• Royal Blue Critical Victory V 
• Navy Blue Active 

And all the other keywords I pulled out were in their own columns. So I now had something I could work with and I started plugging stuff into the Excel template. I put the product name in the “Product” column, the team name in the “Category” column, and filled out the “keyword” columns with some of the keywords I pulled out of the title. I also added in some fields for the full, unedited product name and price.

Next I went to the second page of the Excel file and rewrote all of the sentences. This was by far the hardest part. It can be very tough coming up with 5-10 variations of the same sentence that will all fit together to make a somewhat cohearant product description. I will need to work more on this part but for this test I just wanted to get something out there. You will see that some of my descriptions turned out pretty rough but it is a start.

Once I had those written I went back to the first page of the file and voila! The descriptions were there with my sentences in them. I did need to go in a change the concacation formula a bit to make the description make a bit more sense, include some things I wanted to include like the price and so on. If you are curious here is what the formula looks like:

=CONCATENATE((INDEX(Sentences!$C$4:$G$4,1,TRUNC((5*RAND())+1))),” “,(INDEX(Sentences!$C$6:$G$6,1,TRUNC((5*RAND())+1))),” “,(INDEX(Sentences!$C$5:$G$5,1,TRUNC((5*RAND())+1))),” “,(CONCATENATE(“This sweatshirt is a “,C2,” “,D2,” and comes in “,E2,” sizes. “)),” “,(INDEX(Sentences!$H$5:$L$5,1,TRUNC((5*RAND())+1))),” “,(CONCATENATE(“Manufactured by “,F2,”. “)),(CONCATENATE(“Order your “,G2,” today for only $”,H2,”. “)),(CONCATENATE(“Show your love for the “,B2,” and show off your “,G2,” to all your friends. “)),(INDEX(Sentences!$C$7:$G$7,1,TRUNC((5*RAND())+1))))

Some of the text is the same on every description but there are 4 different random sentences in each description. Overall that is pretty cool. Here are a couple examples of a full product name and the resulting description:

Reebok Seattle Seahawks Royal Blue Vintage Team Logo Pullover Hoodie Sweatshirt: The fall is coming and the weather is turning cold. Stay warm with your Seattle Seahawks Royal Blue Vintage Team Logo   sweatshirt. Whether you’re a Seattle Seahawks fan yourself or shopping for a loved one, we’re here to help. Don’t freeze this winter and stay warm and cozy with a  new Seattle Seahawks hoodie or sweatshirt. This sweatshirt is a pullover hoodie and comes in mens sizes.  The NFL season is underway so gear up today! Manufactured by Reebok. Order your Reebok Seattle Seahawks Royal Blue Vintage Team Logo Pullover Hoodie Sweatshirt today for only $74.95. Show your love for the Seattle Seahawks and show off your Reebok Seattle Seahawks Royal Blue Vintage Team Logo Pullover Hoodie Sweatshirt to all your friends. Don’t wait! Order your Seattle Seahawks Royal Blue Vintage Team Logo   hoodie sweatshirt today.

Mitchell & Ness Chicago Bears Navy Blue Standing Room Premium Full Zip Hoody Jacket: Find the right Chicago Bears sweatshirt for everyone on your list. We have over 1,500 hoodies and sweatshirts available. Whether you’re a Chicago Bears fan yourself or shopping for a loved one, we’re here to help. Don’t freeze this winter and stay warm and cozy with a  new Chicago Bears hoodie or sweatshirt. This sweatshirt is a full zip hoodie and comes in mens sizes.  Shipping is only $4.99 or less per order. Manufactured by Mitchell & Ness. Order your Mitchell & Ness Chicago Bears Navy Blue Standing Room Premium Full Zip Hoody Jacket today for only $119.95. Show your love for the Chicago Bears and show off your Mitchell & Ness Chicago Bears Navy Blue Standing Room Premium Full Zip Hoody Jacket to all your friends. Don’t wait! Order your Chicago Bears Navy Blue Standing Room Premium Jacket hoodie sweatshirt today.

If you want to see more samples you can visit http://nfl-football-jerseys.com/nfl-sweatshirts/

I cut and pasted the new descriptions back into the datafeed, loaded the datafeed up into CSVPig, and rebuilt all the products.

As you can see they aren’t perfect and I need to do a lot more work perfecting how everything fits together but for the most part this is a pretty good start and I can see a ton of potential here. From start to finish this took me about 2 hours and I know future attempts will take less time.

What would I do differently next time?

1. Make sure all the fields are complete. When I pulled the manufacturer out of the title there were some products that didn’t have the manufacturer listed so some of the descriptions have “Manufactured by ” insted of “Manufactured by Reebok” or whatever. So even if I have to just add some generic text in to fill everything out it is better than leaving it blank.
2. Spend a lot more time on writing the sentences. Yeah, I have a bunch of descriptions now but they don’t read well and I think they ended up with too much keyword repetition. And overall they look pretty ugly. I fell into the trap of putting the product name in nearly every sentence I wrote not thinking about the end result and that if I use it in all 5 sentences it will show up 5 times. Pretty obvious now that I look back on it.
3. I want to put a URL somewhere in the description so people can just click and buy from the description. Maybe make some paragraph breaks as well. I really am not sure how to do this with this Excel formula so that would be some trial and error.
4. Spend more time figuring out what keywords to pull out of the product name

But overall I really love this method and I can see myself using this Excel template a LOT in the future. It will be interesting to watch traffic and rankings to see if I see any change based on having “custom” descriptions vs. stock descriptions.

If I can at least get a start on this like I did then I would guess many of you will be able to run with this and do some pretty amazing things with it. I can see a lot of ways to use this in the future.