From what I’ve gathered, Google’s fears AI primarily because its answer context doesn’t provide clickable advertising opportunities. This lack of advertising opportunity (and a troubling, persistent bias problem) is what led to Google stifling the public release of their own generative AI products, leaving a gaping opportunity for OpenAI.
So what’s got Google so worried… and what will that mean for link building service providers?
We’ll start with the 25 year view (bye bye link building as we know it!) and end with the 5 year view (how we’re currently testing AI in a link building context).
25 Year View
Google’s concern is one of audience – they’re afraid that the 2.4 million people per minute with “search problems” will go elsewhere to solve those problems. And so, as a link builder whose work supports organic search ranking outcomes for my clients, I’m concerned too. Especially for our current service array at the 25 year mark.
Because by 2048 Google as we now know it will only be used by librarians and other wonderfully nerdy types who prefer what will essentially be the granular “source code” of the knowledge graph.
Librarians and other nerds are still an influential audience, but they’re also the least convinced by our link-buoyed SEO hellscape and hardly a large enough audience to justify an SEO budget much less an SEOx$1000 PPC budget. Poof. Gone.
Over the next 25 years we’ll begin to see the millions of current use cases for search become slowly, gradually eaten up by AI. And we probably won’t even realize it’s happening.
It will be slow at first, small steps made out to be giant leaps in the press, like in 1880 when Wabbash, IN turned on its electric street lights.
But in 25 years every app, kiosk, car, social network, smart appliance, IOT bicycle or VR human touchpoint will have embedded AI that quietly supports our decisions with how-to’s, advice, suggestions, recommendations and guidance. All the stuff that we used to go to Google to learn…
So where does Citation Labs fit, 25 years from now?
Well we’re going to have to:
- Note where and how search audiences are being better served (and/or cordoned off) by AI.
- Understand user intent – why and how is an AI answer better… and for whom?
- Understand AI algorithms and data sources.
- Learn which signals AI is using as it constructs its recommendations and guidance.
- Create these signals on behalf of clients who want to be included, mentioned, suggested and otherwise cited by AI in whatever the new context may be.
So as everything changes dramatically, in a sense, the bias-injecting function of link building won’t be incredibly different. Its tactics will have to change but if you can affect the data set’s textual source, you can affect the AI’s output.
Clients large and small will want to be included in the suggestions made by AI as it presents across all their audiences’ touch points and so there will still be massive amounts of work that’s needed. Phew, ok. Lots of learning to do but we’ll still have a job as the SERPs eventually take the place of the phone book.
>> 5 Year View
By the end of 2028 we’ll really start to feel AI’s impact on Google’s current dominance – especially given that of their 3.7 billion daily searches, 60% are mobile (mobile’s the easiest and most obvious place to supply AI-supported answers). I hope it will be 10 years so that we have time to figure out how to influence AI outputs, but I think 5 years is more realistic given how far along we already are. Yep, we’re feeling a bit like horse grooms in 1905, watching a Model T splutter noisily past… Better learn to work on engines!
>> 4 Year, 364 Days View
But until that turning point where search users really start to decline, we’re super stoked, as link building service providers, about AI’s capabilities and potential impact on our efforts.
- ChatGPT is Currently Incomplete vs. Actual, Contextual Expertise
We have a process we call “knowledge extractions” in which we select and read from 10-50 documents and extract all the individual units of advice.
We do this to know the full scope of what’s currently being said on a given topic, such as “Yoga Advice for Beginners.” We further rank the advice based on its frequency of recommendation as well as its location within a tip set. Basic yoga advice is a concept for which ChatGPT really shines, so we decided to run a comparison between what we can extract from published “SMEs” vs. what ChatGPT actually recommends.
We saw that ChatGPT had about 50% overlap – but that the extracted knowledge from published SMEs lacked some of ChatGPT’s suggestions too. Our anecdotal takeaway though was that currently there’s still a great deal missing from ChatGPT’s knowledge model and that it still requires a great deal of oversight to ensure it’s actually being useful.
- Content-Type Variants are Easier and Faster to Make and Test for Placements
We’ve long talked about pitching non-article content types to publishers. Stuff like glossaries or FAQs. We haven’t done it consistently, but we did learn recently that we can get ChatGPT to feed back to us in a glossary format.
What other formats could we explore? No seriously, what would you suggest we build and try pitching? I’m at Garrett@CitationLabs.com.
- Local Context and Flavor is More Scalable, at Least for Larger Cities
We’ve found that AI is actually fairly good at infusing content with local flavor such as things to do, great restaurants etc. While it’s unlikely to appeal to a local, we’re not trying to appeal to locals, we’re trying to inform visitors and other newcomers to an area. So it’s perfect for scaling a “local feel” to any content we’re trying to place with local publishers.
We haven’t yet explored at what population size do the local suggestions become less usable. That would be a neat thing to examine.
- SME Quotes are Easier to Infuse into Placeable Content
We have a client with a prominent, quotable spokesperson. This spokesperson is unavailable to us for interviewing. They’re also not quite prominent enough for AI to emulate their voice yet. So we hand extracted and tagged 749 of their most-frequently-quoted-quotes and identified those that we could use within the structure of a larger piece on our needed topic. It was this tagging and structuring part that made us think we could connect it with ChatGPT – since we know how the quotes can frame the flow of an article we can auto-fill with AI and edit with humans.
- Testing if AI Text with Outbound Links Has Ranking Impact Difference
Most rational SEOs I read believe that if a given bit of content solves a search problem better than anything else then who cares if its AI made or not? I have the same opinion, and suspect Google’s algorithm “feels” the same way.
But that said, has anyone noticed anything in the Yandex leak about links from generative AI content? Kidding only kind of. If I were a ranking algorithm would I value links from AI-created content as the same as links from human-created? If that content solved a searcher problem then… yes? I might favor data preference signals from human editors if I could get them, but if this linking content is useful, and its assertions are supported by cited resources… Game on.
We’ll let you know what we learn though.
AI as Channel
Hey digital marketing… Move over there with print, outdoor and direct mail… It’s AIO time (artificial intelligence optimization)! Google’s the phone book and AI’s the new search engine. Not today, not next year. But it’s coming. For certain… as innovative developers apply AI to delivering answers, guidance and other decision support to audiences in action who are so deeply underserved by a list of results.
So. Ask your link agency. Ask your SEO agency. Ask your content agency. Not IF they are using generative AI… but HOW are they testing generative AI in their processes. How are they ensuring that your solution is the one that AI recommends? Because if they’re not – or if they won’t talk about what they’re learning – then you’re already stuck in the past, eking out fractional improvements on the speed of your horse as the cars zoom past.