The complete (and highly subjective) guide to auto applied recommendations


Released towards the back end of 2020, one of the latest products in Google’s machine learning and automation toolkit, Auto-Applied Recommendations takes the AI-based suggestions from the Recommendations tab in the Google Ads interface and automatically applies them to your account.

Google has provided a new interface where you can opt-in or out of AAR and the 36 components that make up the AI-based account improvements, and it’s clear from first glance that the various options are suited differently to different accounts, clients, and management levels.

Now that we’ve had some time to test AAR, and take a deeper dive on the recommendations tab, I sat down with some of the Journey Further Paid Media team and took a closer look at all 36 of the options within Google’s Auto-Applied Recommendations, and their potential benefits and pitfalls, so that you can make a more informed decision as to which of the options would work best for you, and which carry a certain level of risk.


Upgrade your conversion tracking with data-driven attribution

Harry: “The more data you can feed the machine, the higher the chance it will perform. If you have enough data available — and more importantly trust its accuracy — then using DDA seems like the sensible move.”

Marc: “Having said that, this would be a one off change and one that should have already been discussed with clients, so no need to switch this on through AAR.”

Bid more efficiently with Enhanced CPC

Harry: “If there is data available to use enhanced CPCs, such as conversions or revenue, you should use a return-based strategy instead.”

Bid more efficiently with Target impression share

Catherine: “Whilst this strategy is useful for Brand campaigns, I think it’s something we should still be setting up and deciding on the conditions (target, max limit etc.) as well as where we want to be using it.”

Bid more efficiently with Target CPA

Catherine: “With all the bid strategy recommendations, I don’t think it’d ever be something we’d opt into being automatically applied because it’s something we will have likely already considered and tested, and want to have a say in. I think it’s worth reviewing the recommendations on this, as Google may suggest something we haven’t thought about and may want to test, but I can’t think of a case when I’d ever be happy for the machine to make this strategy decision for me and set it’s own parameters. At the end of the day, Google will try whatever it can to make the most money, so I don’t know that I’d trust it to decide what was in my client’s best interest, without knowing what kind of volumes/efficiencies we were trying to achieve.”

Bid more efficiently with Target ROAS

Steve H: “This should generally be the best strategy for retail accounts so I wouldn’t think we’d want to switch away from it often.”

Bid more efficiently with Maximize clicks

Harry: “Similar to target impression share, there are very few situations when you would want to optimise for something as top-level as clicks.”

Steve H: “We’re currently trialling this on a couple of clients where tracking is poor, but we still want to get visibility and traffic for certain products.”

Bid more efficiently with Maximize conversions / Bid more efficiently with maximised conversion value

Scott: “We’ve seen some interesting results from the maximise conversions bidding strategy in the past, but more often than not a client will want more control over efficiency levels. As with all bid strategies, I think we’re best putting some thought into these strategies ourselves, and then ‘steering’ the automated bidding, rather than allowing the AI to select the bid strategy from the outset.”

Adjust your CPA targets / Adjust ROAS targets

Harry: “CPA targets, ideally, should be set based on what a non-revenue conversion is worth to the business. There may be little harm in automatically decreasing this to make the account more efficient, but I wouldn’t want Google increasing the CPA target.”

Catherine: “Not sure that we should be allowing the machine to decide on our targets, but I’d be interested to hear how this gets on — if it were to learn what we regularly change it from and to and then only adjust within that kind of range, it could be useful for days where it thinks it could push harder. That being said, sometimes these decisions are based on the budget we are given by a client, which the system will not be aware of (particularly frustrating when performance is great but the client has to pull back, smart bidding just won’t understand).”

Marginal ROI — Target ROAS lowering

Steve H: “This is one that is worth testing to understand just how smart the automated bidding is with your data — is it good enough to know that more margin can be made by decreasing tROAS?”

Ads and extensions

Use optimized ad rotation

Scott: “Our automated bidding depth, as an agency, is over 80% — so this setting wouldn’t have much impact as automated bidding essentially forces an optimized ad rotation.”

Hannah: “Even on manual bidding, you can’t really A/B test through ad rotation anymore, so there is no harm in using this.”

Test new ad text for repeatedly used phrases

Steve H: “Could potentially be an option for specific clients — where they’re not too protective over messaging and just want the best results.”

Nyall: “I agree with Steve — it would be good to see if this has an impact on Quality Score overtime for less stringent clients.”

Add callouts to your ads

Scott: “A serious thing to consider here is compliance — if you have complete free reign on what you can show in ads, then this may be an interesting one to test.”

Add sitelink extensions to your ads

Catherine: “As Scott said, compliance is my worry here. I’ve had a few instances in the past of the ‘auto extensions’ causing issues when clients see them and aren’t happy with them. Certain clients, for instance, want to sign off on all their ad copy for FCA compliance reasons. I think I’d rather write my own in all instances.”

Add descriptions to your sitelinks

Catherine: “As with all the extensions, it’s going to depend on the industry the client is in in terms of whether there are any compliance restrictions etc. I would, however, rather test this than go straight to allowing it to create new ads, which could be more problematic. If we have any gaps and this will boost quality score, then maybe not too much can go wrong given how many characters it can go wild with in things like callouts, it just needs to become part of our task list to review what has been added to check for any mistakes.”

Add structured snippet extensions to your ads

Scott: “These are probably the safest creative element that you can let machine learning pull together for you. As structured snippets fall into specific categories, and the services/products themselves should come from site, you’re probably better off including these than not! (Unless on a particularly compliance-strict client!)”

Create new versions of your ads

Catherine: “Once again, compliance is my worry here, but I would assume that machine learning would be trying to combine different versions of the best performing ads, rather than coming up with it’s own creative so I would hope nothing would go too awry. Another worry I have is that we could end up with tens of variations of ads within the account which makes it harder to report on what is actually working, being unable to group any of them together?”

Create Dynamic Search Ads

Catherine: “Something I think we would definitely want control over in terms of picking the landing pages and copy etc., as well as where it fits within our structure and naming convention etc. It’s another one that would probably be very useful for small businesses who don’t have the time to give accounts the attention they need, but as an agency I think DSA should be a conscious decision to implement.”

Add responsive search ads

Scott: “Given the strong results we’ve seen from Responsive Search Ads, mainly through the additional auctions they appear in, and therefore the extra impressions and traffic they drive — no account should really be without them. Again though, this comes down to a question of compliance and writing the best possible copy for your clients. This should be performance driven.”

Add responsive display ads

Jack: “With studies stating 60–80% of ad performance is based on creative, we’d argue that it could be more important than targeting our media spend to achieve success. At Journey Further, we have a performance-driven creative team, so wouldn’t rely on Google’s automated ‘one size fits all’ approach.”

Add assets to your responsive search ads

Scott: “The struggle with RSAs, and adding new assets is determining the success of these at scale — through the interface you can only see how a single asset is working on a single ad, which may mean it’s difficult to see if this option is working — you can’t see if the asset it’s added has been a success.

Having said that, Google have now followed the plan laid out by Microsoft and are starting to offer cross-campaign asset reporting very soon!”

Improve your responsive search ads / Improve your responsive display ads

Catherine: “Same as ‘create new versions of your ads’ — could be worth a test to see what it does, but could end up creating quite an untidy account. I’d rather be able to see some suggestions and choose what to do with them.”

Keywords and targeting

Expand your reach with Google search partners

Scott: “The time saving element of this one is questionable given a view of search partners performance is really quick to pull together. The performance implications of this one can be fairly significant depending how your ads perform across search partners versus search.”

Harry: “Enabling search partners is the fastest way to spend more without seeing a return. Good for Google, bad for the client. If search partners are disabled, it’s for a reason.”

Marc: “Search Partners can be a great way to expand your reach and drive more volume within your account, however the results usually vary by client and vertical. This may well have been something we have tested in the past and made a conscious decision to turn off. By following the Google view of always having this on could be detrimental to certain clients.”

Remove redundant keywords

Catherine: “What is classified as ‘redundant’? There could perhaps be things in the account that are more seasonal, so could be judged as redundant depending on the time of year etc. — not something I’d want to opt in without understanding the rationale behind it, and not really sure of the benefit to performance in terms of the eyes of machine learning”.

Remove non-serving keywords

Scott: “This is fine, if you’re looking to keep a tight and tidy account — however, it does raise the question; why are those keywords not serving? Is there something that could/should be done differently?”

Pause poorly performing keywords

Steve H: “You could maybe lose track of what’s live in your account, and this could be a problem for any trend based or stock influenced KW’s — shouldn’t smart bidding optimise spend away from areas that don’t perform anyway?”

Harry: “I would be weary to automatically implement this, as ‘poorly performing’ can often be subjective. As we are constantly delving into the detail of our client’s accounts, the only poorly performing terms to identify would be any non-directly returning keywords clients have specifically requested, such as bidding on a competitor with the aim of stealing market share.”

Marc: “The concept of pausing poorly performing keywords sounds like a no brainer, however we need to be careful that if we want to remove certain terms that they don’t start appearing elsewhere in the account via broad match or DSA.”

Add new keywords

Scott: “Given some of the examples I’ve seen in the recommendations tab, I’d stay away from this at all costs!”

Catherine: “I agree! Yet to see anything good in these recommendations, and I’d hope we’d be mining our own new keywords from SQRs & even DSA campaigns before resorting to opting in here.”

Steve H: “Potentially the most valuable recommendation? I can see it messing with account structures though, if it worked perfectly though something like this could be of enormous benefit to certain clients.”

Marc: “New keywords being identified could be very useful, however I wouldn’t want them auto applied as it could interfere with the account structure, it would be better to take the list of suggested terms and build them into the account manually.”

Add phrase or broad match versions of your keywords

Steve B: “Interesting option, given that Google is moving away from Phrase Match…”

Add negative keywords

Catherine: “Guessing this would just be based on things like poor CTRs and landing page relevance, could be problematic for new trends etc. if it’s adding in things that it’s deeming irrelevant, but they could be relevant if we knew about them … thinking particularly for fashion where new search terms pop up all the time and are sometimes useful to feedback to client for buying decisions.”

Remove conflicting negative keywords

Steve B: “Should it remove the negative keyword, or the positive one? This should only trigger if something’s gone wrong, and either response may be appropriate.”

Add audience reporting

Scott: “As this just adds new audiences set to Observation, this will have no impact on actual performance, and so is the safest auto-applied recommendation to use on even the most technically sound account — any additional insight in your reporting can only be a good thing!”

Adjust your keyword match versions

Steve H: “Danger!”

Catherine: “I’d have thought a more useful tactic would be to implement a new keyword with a different match type — looking back at a keywords performance could be very misleading should the match type be changed at some point.”

Broad match for fully automated conversion-based bidding

Steve H: “We’re currently doing a lot of testing around this, and it’s not too much of a stretch to think this could potentially be our default approach in the future?”

Scott: “I agree — given the significant changes Google are making at the moment around automation and value based bidding, I can certainly see this being something they move towards.”

Display targeting expansion

Jack: “Opportunities do exist within growing markets, identifying these can help extend reach and grow awareness and conversion actions. I tend to use the keyword planner tool to better identify user intent and further expand our audiences so that I don’t lose control of who we’re putting our ads in front of.”


Fix your ad text

Scott: “This is the only recommendation under the “Repairs” banner in auto applied recommendations, and as far as I can see, simply fixes basic errors like misspellings. I’d like to think our internal review process would prevent errors like this from happening, but it can’t hurt to have a helping hand.”

In Summary

Marc: “There will be some suggestions that are very useful in AAR, however it will take someone manually reviewing them to dig out the good ones.

You could potentially turn AAR on and have positive improvements for months, however I feel that you are only one bad change away from spending a significant amount of budget on something that you don’t want to, which would be a bigger loss than the rest of the gains combined”.

Catherine: “The proposed benefits of AAR to small businesses running their own activity are clear to see, with Google implementing best practice and taking care of things that would be time consuming for someone who has a whole host of things to manage.

For agencies however, some of the recommendations suggested may totally contradict the strategy that has purposely been put in place — there is no way machine learning can appreciate the context of why things are the way they are, and a one size fits all approach is not suitable.

I think there is still a lack of visibility as to what some of these changes actually do, and how the decision is made by Google to apply them, so we need to tread cautiously with opting in to recommendations. We need to ensure we are minimising risk and only taking up options where we know the outcome will not be detrimental to performance, letting machine learning step in to save us time and spot opportunities that we can’t, just how we have with automated bidding.”

Nyall: “It’s easy to see that Google is trying to get advertisers to feed the bidding algorithms as much data as possible for the models to achieve a positive result according to its objective overtime, but we as pilots should be doing this in line with our clients strategy (within a range that is efficient when paired with the clients KPI and the data presented to us), rather than relying on a tick box exercise that has many caveats and possible detrimental effects to account efficiencies in the short term.

I’ve seen a lot of negative discussions around how Google is trying to take away control from accounts and advertisers, but the changes are being put in place with the best interests of advertiser’s performance in mind. For the time being, I don’t think the technology is there yet, which is why our roles as ‘PPC pilots’ is more necessary than ever until that time comes.”