Flexible Headlines

[David Repas]

Okay, here's the basic idea.

Flexible Headlines would be a service that allows content creators (news orgs, blogs, etc.) to test multiple headline variations for each story that is posted.

After a certain period of time (or number of views), the number of headline variations served would begin to narrow (least popular would disappear) - and at some point the article would settle on the most popular headline.

Obviously, if you can substantially increase click-through rates to news stories, blog posts, etc. based solely on the headline used - you are creating real value for content creators who typically rely on page views and in turn advertising revenue.

[Nicholas Holland]

*Whistles...* pretty slick idea. You thinking an API / Javascript controlled headline? Or would the entire article be served from your platform? Why not A/B articles?

[David Repas]

I'm thinking that even though a service like this "could" be used by really anyone, larger news orgs, blogs, etc. would be the target since they depend on traffic for ad revenue. More traffic = more ad revenue = a willingness to pay us if we make them more money.

At first, I was thinking of using a Javascript powered headline, but I do have concerns over how seamless it would be with existing CMS. All of the larger content creators surely use an existing CMS to manage content (or maybe something like Wordpress).

So, the question is how can you seamlessly integrate a service like this into existing applications? I'm not entirely sure. Maybe start with some of the more popular platforms? Like Wordpress, Dupal, etc.? and build plug-ins for them?

Or maybe just build the infrastructure, APIs, tools, etc., and offer it to anyone to tap into - in other words, have the community build the plug-ins for various platforms. Then make money by offering premium services, like a headline recommendation engine, advanced analytics, etc.

Any thoughts?

[Nicholas Holland]

My gut leans towards a plug-in. I can't think of enough 'additional' features to merit a service rather than a single plugin. However, this plug-in could be commercial and you might do well selling a few of them :)

[David Repas]

I wonder if there's some sort of concept similar to Google AdSense (but for content) that could be explored?

The basic idea would be that instead of sites serving targeted text-based ads (powered by Google AdSense), sites could serve targeted text-based content headlines.

Content sites (such as TechCrunch) would pay-per-click to their articles. But what would make this even more interesting than just gaining traffic is that content sites would also be able to do headline variation testing. And then, when all is said and done, use the best performing headline in the final version of the article. The version that lives forever on the web, is indexed by search engines, etc.

This structure would avoid a lot of the pain of merging the Flexible Headlines service into existing platforms (like WordPress). It could be a somewhat independent service.

Just another thought....

[David Ledgerwood]

Couldn't you fabricate this with Google's Website Optimizer? I've haven't used it, but I thought it works with any chunk of code on the site with A/B testing. What it may not do is automate the degradation of the least popular headlines.

[David Repas]

It definitely has some similarities and may be something to explore. From what I know about Google's Website Optimizer, I think its main goal is to increase conversion once a person lands on a page. So, for example, what headlines or page design converts the most users into buying or signing up, or whatever.

I'm starting to think that the most valuable use for Flexible Headlines would be to determine what headline will drive the most traffic to an article. So, as that article is distributed via RSS or search engine links or whatever, a content creator can be sure that they are using a headline that will optimize click-through rates over time.

[Michael Summar]

I'll pile onto Ledgerwood's advice that this is being done like crazy on millions of pages using tools such as Omniture Test&Target, GWO, SiteSpect, and Optimost. Although, I think you are running down a great path (it's the space I play in). My 2 cents ... take it a step further by having the delivery be true behavioral targeting (BT) versus an A/B, one gets more "success" events than another. Gather several different data points such as referrer, daypart, previous visit history, previous success events, etc., to behaviorally target the user with the appropriate ad. Build a robust enough tool that uses scientific methods for the behavioral targeting ... and you will definitely have a market. All the other BT tools on the market are outrageously expensive and out of reach of virtually everyone outside Fortune 100 companies.

[Justin Davis]

Interesting idea. I'm definitely a fan of A/B testing in web scenarios, and I think it provides a lot of valuable information when testing designs.

A challenge I see is the short effective life-span of an article headline. Headlines are pretty disposable - like an ad, they're designed to interrupt a user/reader long enough to get them through the first paragraph. A/B testing a headline may be a great idea, but I'd worry that by time you got measurable results, the largest opportunity for optimizing that headline has been lost. That initial wave of readership, the spike of traffic that happens directly after an article is published, will never see the optimized version (nor do they need to care).

The "long-tail" argument is interesting. Since these articles live forever on the web, making a headline as effective as possible makes sense. That said, I'd argue that articles and content consumed in the long-tail aren't spread primarily through headlines, they're spread through actual body content. Per the long-tail theory, content in that part of the curve is accessed by few people, consistently over time - niche content/products. I'm not sure that content spreads in that part of the curve the same way it does in the beginning part, where headlines (read: advertisements) really matter.

[David Repas]

Lots of good points.

I definitely agree with the concern that optimizing a headline during the initial wave of readership could be problematic. In the ideal world, a content creator would use an "optimized" headline from the moment the article is posted, sent out via RSS, etc.

But, how could one do that?

Just brainstorming here... Maybe create a service to test headline variations before the article is actually fully released to the general public? Something along the lines of pre-releasing content to a small group of users. Or creating the tools that allow sites to pre-release content (in a controlled and limited way) and then learn from the interactions with that content.

Any other ideas?

[Michael Summar]

I like the way Justin thinks. In the world of media planning and analytics, the lifespan of an ad with the breakpoint of diminishing returns is able to be calculated. This is usually done in the offline world. Perhaps the headline idea morphs into a live engine constantly delivering the optimal content (headline, ad, etc.) making use of the breakpoint.

David Repas said:

Lots of good points.

I definitely agree with the concern that optimizing a headline during the initial wave of readership could be problematic. In the ideal world, a content creator would use an "optimized" headline from the moment the article is posted, sent out via RSS, etc.

But, how could one do that?

Just brainstorming here... Maybe create a service to test headline variations before the article is actually fully released to the general public? Something along the lines of pre-releasing content to a small group of users. Or creating the tools that allow sites to pre-release content (in a controlled and limited way) and then learn from the interactions with that content.

Any other ideas?

[Kate O']

Hey, how did I miss a great discussion on A/B testing and targeting? Wow. :)

Like Summar and Davis have said, this is a great idea but there are tools that are doing really similar things, and the timeline of a headline's effectiveness makes this challenging.

On the other hand, this is a whole lot like testing subject lines in email. The point of that testing isn't just to optimize the email being tested; it's to inform the process in an ongoing way so that the following emails can start out from a more optimized point.

In other words, maybe what this is is more of a methodology that can be built into content authoring tools, like MarsEdit and other blogging apps. Maybe one starts with a hypothesis that a particular headline variant will appeal to a certain subset of the audience while another headline variant might appeal to another subset, and one then sets up a whole bunch of segments to test for. Once you have results that suggest that, say, the snarky headline worked well with audiences who were referred from Slashdot, but the "just the facts"-style headline worked best with audiences coming from Google, you can probably optimize future content distribution for best chance of clickthrough.

[Jake Robinson]

Hey Michael,
is what you're describing anything like Multi-Variant Testing?

Michael Summar said:

I'll pile onto Ledgerwood's advice that this is being done like crazy on millions of pages using tools such as Omniture Test&Target, GWO, SiteSpect, and Optimost. Although, I think you are running down a great path (it's the space I play in). My 2 cents ... take it a step further by having the delivery be true behavioral targeting (BT) versus an A/B, one gets more "success" events than another. Gather several different data points such as referrer, daypart, previous visit history, previous success events, etc., to behaviorally target the user with the appropriate ad. Build a robust enough tool that uses scientific methods for the behavioral targeting ... and you will definitely have a market. All the other BT tools on the market are outrageously expensive and out of reach of virtually everyone outside Fortune 100 companies.