Big Horn, Small Cashews: How to Select an Agency that Understands AI Search

Imagine you are doing a quarterly review, and the graph is showing a downward trend. Before you even ask for an explanation, your agency is prepared to give one: "It's AI. Everyone's traffic is down."

They're not wrong. That's actually the frustrating part. Then someone on your team mentions they asked ChatGPT which companies handle what you do, just out of curiosity. Your biggest competitor came up first. You didn't come up at all. And nobody in the room has a plan for that, because nothing in the agency's report measures it.

That is what this guide is about. Search has changed. People started asking their AI assistants questions and for shortlists, and very few of them actually clicked through to any of the agencies' websites. Most of the agencies were aware. Very few changed their ways of doing things. Quite a number simply included the word 'AI' in their presentation. At Pressfit, we have a name for those agencies: big horn, small cashews. Very loud about AI. Not much inside the shell.

This guide is meant for those who are currently in the process of choosing a marketing agency or secretly questioning whether the one they’re using is up to date. Here, you will find the questions to distinguish a skill from a brand makeover, how to recognize a good answer from a bad one, and a simple tool to test your AI readiness even before the meeting happens. All of this is based on our proprietary research at Pressfit, which consists of more than 150,000 citation records and 24,000+ actual customer inquiries via ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews.

What Actually Changed When People Stopped Clicking?

People still search. They just don't leave the search page nearly as often as they used to.

SparkToro looked at US Google searches from January through April 2026, using Similarweb clickstream data, and found that 68% of them ended without a single click to any website. In 2024, that number was about 60%. And the study didn't even include searches made inside the Google app, where people probably click even less.

That is partly due to the importance of AI Overviews. Depending on which tracking source you use, they currently make an appearance in 20% to 45% of all Google searches. If they appear, people will read the summary and continue. A study by Pew Research Center done in 2025 discovered that if there was an AI overview available, the user clicked on one of its sources only 1% of the time.

This does not even take into account what happens completely outside of Google. Consumers request a comparison between vendors from ChatGPT. They request a shortlist from Perplexity. They request an explanation of the differences between three tools mentioned by a colleague from Claude. This is not visible in your analytics.

The old buyer path looked like this:

Search → Click → Visit → Convert

The new one looks more like this:

Question → AI answer → Shortlist → More research → Branded search → Pipeline

Look at where your website sits in the second one. Near the end, if it shows up at all. Pressfit's client work points the same way. In most of the buyer journeys we review, the brand's own website accounts for only 20 to 40% of the research a buyer actually does. The rest happens in AI answers, review sites, Reddit, YouTube and creator content.

So "traffic is down" can't be where an agency's explanation stops. Traffic was always standing in for a bigger question: are the right buyers finding you and considering you? That question still matters as much as ever. You just have to measure it in different places now.

If you want the longer version of this argument, read Zero Clicks: Why Brand Visibility Beats Traffic in the AI Era.

Why Isn't Getting Cited the Same as Getting Chosen?

This is the idea most agencies haven't caught up with yet, and honestly, it's the most important one in this guide.

When an AI tool answers a question, it does two separate things. It pulls information from sources, which is a citation. And sometimes it names specific companies as good options, which is a recommendation. Those two things overlap a lot less than you'd expect.

Lily Ray, one of the most respected SEO researchers working today, tested this between April and June 2026. She tracked 100 B2B "best [category] software" searches in Google AI Overviews. A lot of those Overviews cited listicles that brands had written about their own category. You know the type: "The 10 Best CRM Tools," written by a CRM company, with its own product sitting at number one.

Those self promotional listicles were cited 323 times. In 224 of those cases, 69% of the time, Google used the brand's article as a source and then recommended somebody else.

One example from her study really sticks. An LMS company called Oasis LMS had its "best LMS for selling courses" article cited in the AI Overview. The brands Google actually recommended were Kajabi, Thinkific, LearnWorlds and Teachable. Every one of them was mentioned in Oasis's own article. Oasis basically wrote the comparison that sent buyers to its competitors.

The one self promoter that did get recommended for "best CRM for small business" was Monday.com, a brand with a domain rating in the 90s and around 75,000 referring domains. In other words, a brand the rest of the internet already vouched for.

That's the lesson. AI tools are happy to use your content as evidence. Whether they choose you depends on what everyone else says about you.

And this is exactly why it matters when you pick an agency. An agency that only reports citations can make you look like you're winning while your competitors collect the recommendations. "You were cited 40 times this month" sounds great in a report. It means very little if the answer then tells the buyer to go with someone else.

We go deeper into this in why your "best of" article gets cited in AI search but recommends your competitor instead.

Why Should You Treat AI Search as Five Audiences Instead of One?

A lot of agencies talk about "AI search" like it's one channel. It isn't, and Pressfit's research shows just how different the platforms are.

Across our 150,000+ citation records, 75.8% of the domains that got cited showed up on only one of the five platforms. Another 13.6% showed up on two. Only about 1.3% were cited on all five. So a source ChatGPT trusts is, most of the time, a source Perplexity and Google AI Overviews never touch.

The platforms also cite at very different rates:

Claude gives out fewer than five source slots per answer. Google AI Overviews give out about fifteen. That changes how hard each one is to break into and what kind of content tends to get picked.

Then there's one last bit of complication. You could ask an AI the same question twice and receive two different answers. This was noted by SEO specialist Ethan Lazuk, among others, and anyone keeping track of their own AI presence will run into it very quickly. This is why Pressfit looks at citation share when asking the same buyer question multiple times.

Put all of that together, and you get a simple rule for judging agencies. One screenshot from one tool proves almost nothing. It's one answer, from one engine, on one day. If that's the evidence an agency brings to the meeting, they're showing you a lucky moment, not a strategy.

What Questions Expose Whether an Agency Really Understands AI Search?

Most "how to choose an agency" articles hand you a list of qualities.

• Data-driven

• Adaptable

• Strategic

Every agency on earth will tell you they're all three, and they'll mean it.

So instead, here are seven questions. For each one, you'll see why it matters, what a real answer sounds like, and what a bluff sounds like. You can take these straight into your next agency meeting.

1. "Can you show me where we stand right now?"

Before anyone can improve your AI visibility, they have to measure it. That sounds obvious, but a surprising number of agencies skip this step and jump straight to recommendations.

A real answer starts with a fixed set of questions your buyers actually ask. Not keywords. Questions, like "what's the best payroll software for a 50-person company" or "which cybersecurity firms work with hospitals." The agency runs those questions across all five major AI platforms, more than once, and tells you how often you show up, how often your competitors show up, and whether you're being cited, mentioned or actually recommended. That becomes your baseline, and everything after gets measured against it.

The bluff seems to be an isolated screenshot from ChatGPT talking about your profile being mentioned. Or a "hey, you've got a good AI footprint here" with no stats attached to back it up.

When you go into the meeting, ask them how many queries they track for buyers, on what platforms they do so, and how often they ask the same query. If they don't know all three, they have no metrics in place.

2. "How does your approach vary if the purchaser uses ChatGPT instead of Google?"

If you only get to ask one question, ask this one.

Agencies that have actually adapted get specific fast. They'll explain how ChatGPT, Claude, and Perplexity pull from different sources than Google. They'll explain how they figure out which third-party sites the AI tools trust in your category, and how they get you mentioned there. They'll talk about structuring your pages so an AI tool can lift a clean answer straight out of them. They'll probably admit that some of the old playbook still matters, because many AI answers are built on top of regular search results.

The weak offering is just your 2023 package with new labels. Blog post, backlink, and monthly report are all now termed "AI optimized." If you can't identify one deliverable that's different, there isn't any.

3. "How do you get us from cited to recommended?"

Don't forget the 69%. The easy part is being cited. The hard part is getting chosen.

The right answer is all about providing proof that exists beyond your site. Positive reviews on websites that your customers trust. Mentions in industry publications. Genuine comparison content that shows where you shine, and where you don't. Case studies with solid data. A story that defines your target audience and is consistent across channels. Pressfit's research showed that mentions of brands on independent websites correlated with citation rates better than backlinks, which means that this is not just a bonus step. It is the core of the whole process.

A bluff is a plan to write more listicles that rank you first on your own blog. That's the exact strategy that got cited and passed over 69% of the time.

A good follow-up: ask how they'd figure out why a competitor gets recommended instead of you. A strong agency will talk about checking the sources behind that recommendation, how the competitor is described, and what proof backs up those claims. Then compare all of it with how your own brand shows up.

4. "What will you report every month, and how does it connect to revenue?"

AI visibility is only worth paying for if it moves something you care about. So the reporting has to connect the dots.

Good reporting has a few layers. Citation share, which is how often you're cited across your tracked buyer questions. Recommendation frequency, which is how often the AI actually suggests you. Branded search, because buyers who see you in an AI answer often go search your name later. And pipeline, meaning account activity, demos, and opportunities.

Such an agency will not shy away from being transparent about this. They would simply tell you that "the number of citations increased along with the amount of branded searches during the same time period," without overstating how each citation resulted in a sale. Overstating it is a red flag, because AI-driven influence is difficult to measure with analytics tools.

Watch out for a dashboard full of mention counts with no link to anything commercial. Or the opposite: a very confident claim that their AI work "generated" some specific revenue number.

5. "Who actually does the work?"

Plenty of agencies can find problems. Far fewer can fix them.

AI visibility usually needs hands-on changes. Restructuring pages so the answer comes first. Adding tables and lists. Cleaning up schema markup. Fixing technical issues that stop pages from being crawled. Improving product data if you sell online. Sometimes CRM connections or automation, so the visibility data actually reaches your sales team.

Some of these changes matter more than you'd guess. Pressfit's research found that pages leading with a tight 60- to 80-word answer were cited more often on Claude and ChatGPT. Numbered lists and tables got quoted word for word, while the same information written as regular paragraphs got paraphrased away. Those are implementation jobs. Someone has to actually go in and do them.

So ask who does each of these things, how fast, and who checks the work afterward. A good agency answers that without having to think about it.

A bluff is an audit where every recommendation ends with "your internal team can handle that." If that's the deal, you're not really hiring an agency. You're buying a very expensive to-do list.

6. "Are the fundamentals in place?"

This one sounds boring. It's also where a lot of AI visibility quietly falls apart.

AI tools still need to find, read, and trust your content. That means pages Google can index, a site that loads properly, business information that matches everywhere it appears online, and real customer reviews. Technical SEO consultant Glenn Gabe recently published a breakdown of how to find PDFs that Google has deindexed. If your best research or case studies live in PDFs, they may have dropped out of the index without anyone noticing, and AI tools that lean on Google's index can't cite what isn't there.

Any agency worth hiring will want to run a technical and indexing check before the fancy AI work starts.

A bluff skips straight to "AI visibility" like the foundation doesn't matter. It does. It's still most of what search engines and AI tools use to figure out who you are.

7. "Do your channels talk to each other?"

If you're running paid ads, SEO, content, and social, those teams should be learning from each other.

What a paid campaign shows about which messages people respond to should shape your organic content. What your organic content shows about buyer intent should shape your paid targeting. And what AI answers say about your category should feed into both. Test, measure, adjust, repeat.

Ask them for one actual example, from an actual client, of something one channel learned that changed what another channel did.

What you don't want is five specialists under one logo who don't really talk to each other. Each one might be great at their own job. But nobody closes the loop, and the loop is where most of the improvement comes from.

What Are the Red Flags of a "Big Horn, Small Cashews" Agency?

Some warning signs are easy to spot once you know what to listen for. If you hear any of these, slow down.

• Guaranteed AI rankings. Nobody controls what ChatGPT or Google's AI says. Anyone promising you a permanent spot in AI answers is promising something they simply can't deliver.

• One screenshot as proof. One answer, from one tool, on one day, is not a result. Ask how often it happens across many questions and many runs.

• A link building plan renamed "AI visibility." Backlinks still count, but Pressfit's research found they work more like a tiebreaker than the main lever.

• A listicle factory. If the plan is mostly "best of" posts ranking you first on your own site, scroll back up to the 69%.

• Every citation equals revenue. Attribution in AI search is messy. An agency that makes it sound simple is either guessing or selling.

• The badge instead of the work. Partnership, certification, and flashy AI presentation slides are all good. But they are not evidence. Try asking what the partnership has really done for a client like you.

• No talk about competitors. AI answers are comparisons. If an agency only looks at you and never at who's getting recommended in your place, they're missing half the picture.

How Can You Check Your AI Visibility Yourself Before the First Meeting?

You don't need a paid tool to get a rough picture. An afternoon and a spreadsheet will do, and you'll walk into the agency meeting knowing more than they expect.

1. Note down 20 questions that your customers ask. What would someone looking to solve their problems search for before they even heard your name? "Best [your category] for [your kind of customer]." "How can I fix [your solution]." "Alternative to [competitor]." Your salespeople can provide this information too. This is pure gold.

2. Run each question on ChatGPT, Gemini, Perplexity, Claude, and Google. If possible, create a new search session for yourself and take notes about each response, including whether your company’s name was used, whether your website was referenced, and whether you were recommended.

3. Run your most important questions a second time. Answers change between runs. If you show up once out of three tries, that's worth knowing.

4. Work out your citation share. Divide the number of answers that cite you by the total number of answers, then multiply by 100. Cited in 9 out of 50 answers? That's a citation share of 18%.

5. Check Google Search Console. The performance report for generative AI tells you how many impressions you've gotten from AI Overviews and AI Mode. It's not going to be perfect, but it's actual Google data.

6. Look for patterns. Which competitor shows up most? Which platform ignores you completely? Which sources keep getting cited in your category?

Come to the meeting prepared with that sheet. Ask the agency to explain their reasoning behind those holes and how they would address them. This will say more than any pitch deck ever could.

This works well for a small set of questions. Once you're tracking hundreds of questions across five platforms and several competitors, doing it by hand gets slow, and that's usually when a dedicated tool starts to make sense.

What Should Your Agency Scorecard Look Like?

Take this into your next meeting. Mark each answer as weak or strong, and see where the agency lands.

If an agency gives strong answers on five or more, you're probably talking to people who've actually done this work. Under three, and you're mostly hearing horn.

Where Does Pressfit Fit In?

Pressfit is created based on the challenge presented in this guide. We monitor how well brands perform across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews, identify which brands are being recommended in place of them, and correlate all of that with their performance in branded searches and pipeline because it's not just another metric in the dashboard. And then we make sure that all the holes are plugged, from web architecture to messaging.

If you want to see where you stand today, drop your URL on our get-onboarded page, and we'll run our system against your site, then walk you through how you're showing up in AI search right now.

So, How Do You Pick?

The agencies worth hiring right now aren't always the ones with the best AI pitch. They're the ones who can show you where you stand, explain why your competitors are getting chosen, do the work to change it, and tell you honestly what the numbers mean.

Everyone has their AI slide right now. Ask the seven questions, use your own evidence, and observe who has anything meaningful to say without it.

FAQ