A client spending about $20k a month asked me whether he should buy his way into AI answers or do the slower work of getting cited in them. His paid guy said buy: the campaigns were already running, and he had a pipeline target to hit. His SEO guy said earn: why pay to sit beside an answer that could name you inside it? Smart people disagree because they are solving for different clocks.

I have run the paid side for years, and its case is stronger than the usual organic-first sermon admits. So I will make both arguments at full strength before I pick one. My verdict, for most businesses, is earn the citation first. An ad bought in an answer you are not cited in is paying rent on someone else’s credibility.

Position A: Buy the slot and get to work tomorrow

If I am arguing for paid, I start here: you can enter the test with campaigns you already run. There is no dedicated AI Overviews campaign type to learn and no separate approval queue. Google serves Search, Shopping and Performance Max ads into AI Overviews and AI Mode from existing campaigns through the same auction at the same CPC, though there is no guarantee your ad will show. That last part sounds like a weakness until a sales team needs pipeline this quarter. You can set a budget, retain geo, schedule and bid controls, and read the conversion data your CFO already knows how to use.

Paid also gives you a way to act without waiting for a page to be rewritten, fetched and cited. If a query already brings in customers, you do not have to pause that work while you try to earn a place in the answer. You can keep buying the traffic and judge the campaign against revenue. That is not a glamorous argument. It is a useful one when a deadline is real and a citation is not something anyone can schedule for Friday.

The available inventory is narrower than the pitch makes it sound. An SE Ranking study of 50,032 commercial keywords found text ads on 29.45% of AI Mode queries. CPC predicted visibility better than anything else in that study: 24.33% coverage under $2 CPC versus 53.56% at $10 and up. Ads are also limited to text and Shopping placements in English on mobile and desktop in 12 listed markets, with no ads inside Overviews for sensitive verticals such as finance, healthcare, gambling and politics.

Then there is the control you do not get. Google offers no targeting switch for AI Overviews, no opt-out and no segmented reporting. I used to tell clients that kind of opacity was a dealbreaker. I was wrong. You can still run a paid test through the campaigns you have and read its contribution to revenue, even if you cannot isolate the Overview placement neatly. If the choice is between imperfect visibility now and an unknown citation date, paid has a serious case. Buy when the revenue deadline is closer than the organic one.

Position B: Earn the place inside the answer

If I am arguing for earned, I start where paid is weakest: the ad sits beside the answer; the citation sits inside it. OpenAI calls this answer independence: ads do not influence ChatGPT answers and appear as a labeled card below the response. Google likewise keeps the Sponsored unit separate from the AI summary. What the deck calls “coverage,” I call renting the footnotes while someone else writes the verdict.

The separation is not just visual. Among the keywords they bought in the SE Ranking study, only 11.53% of advertiser domains appeared among cited sources, falling to 1.95% at URL level. About 85% of those advertisers did not rank organically for the same query either. Buying the slot does not buy the citation. It can put your name on the page while the answer draws its authority from someone else’s work.

That matters because the answer helps set the reader’s shortlist. Cited brands see about 35% higher organic CTR than they do at the same position without a citation, and Seer documented up to a 91% paid CTR halo when a user saw a brand in the Overview and then saw its ad. The mechanism is plain enough: the reader meets the brand in the answer before reaching the listings below it. Paid can benefit from that familiarity, but paying alone cannot create it.

Missing the answer can hurt, too. Organic CTR on AI Overview queries drops 15 to 46% overall and 46% on informational queries, with Ahrefs tracking an initial 65% collapse that settled at 30 to 40% below baseline. Those numbers do not mean every citation becomes a lead. They do explain why “we still rank” can sound reassuring while fewer people click. If the answer satisfies part of the search, the sources it names have a better chance of being the ones a reader follows.

Then there is the rent math. Average Search CPC sits at $5.42 versus $2.32 in 2016, a 134% rise, with legal averaging $9.87 while arts sits at $1.63. A citation has no per-click charge. An ad stops bringing in paid traffic when you stop paying for it. Earned visibility is not free to build and it is not guaranteed to stay put, but it does not start a new meter each time someone visits. The earned advocate’s best case is not “SEO costs nothing.” It is this: do work that can keep paying you back without charging for every click.

Where the two arguments break

Paid first, because I know its failure modes by heart. The ad buyer gets reach without residence. The AI Mode study found ads on fewer than a third of the queries it examined, higher CPCs went with greater visibility, and the reporting does not tell you enough about the placement. Impressions inside the Overview report as Top Ads, with no separate split to read. I have watched accounts spend real money next to an answer that cited competitors without getting a clean view of what that placement contributed. When the report rolls up the placement, you can judge the campaign; you cannot inspect every part of the argument you thought you were testing.

The credibility gap makes that limitation sting. If a reader has just seen three other sources named in the answer, your labeled ad has to earn attention from underneath them. You may still get the click. You may even need it this week. But you keep restarting the meter, and higher spend does not change who the answer cites. Do not mistake the ability to buy a placement for control over the answer.

The earned side fails differently. A citation is hard to steer and slower to assess. It takes weeks of rewriting pages, fixing fetch and structure, and waiting to see whether the model notices. Attribution is thin, testing is slow, and nobody can give you a reliable date for a citation. I tell clients who want one the same thing I told them about Quality Score in 2019: I can show you the inputs that correlate; I cannot promise the output on Friday. That is why the paid-search instincts that work in auctions can fail in answers. A bid can enter an auction. It cannot make a paragraph worth citing.

That uncertainty is not an excuse to buy another dashboard and wait. It is a reason to be precise about the pages and queries you choose. Earned work deserves the same commercial discipline as paid work: start where the business already has evidence of demand, then change the page that should answer it. Otherwise you trade a clear CPC bill for a vague content bill. Neither is a strategy.

My verdict: earn first, except when the clock wins

I rule for earn first. The deciding consideration is placement. The citation lives inside the answer the buyer reads as a shortlist; the ad lives in a labeled box beside or beneath it. If you buy that box without earning a place above it, you are paying for clicks on a page where competitors may already have the reader’s trust. I have seen a version of that movie with Quality Score: the advertiser with the better landing page paid $4 while the lazy one paid $9 for the same click. The lesson is not that citations hand you a predictable CPC discount. It is that buying traffic works better when the surrounding experience gives a reader a reason to choose you.

Buy first anyway if you have a revenue date that does not move. A launch next week, a seasonal peak, an idle sales team or a test you need to read by Friday: keep the paid campaigns working and accept the rent. If Google keeps your vertical out of Overview ads, that exception is not available there. If clicks already cost $9 and climb while you have no useful content to support them, buying more is not a fix for what the answer says about your category.

For everyone else, the order is clear. Earn the citation on the queries that decide your category, then buy ads on those same queries to add the paid placement. You do not have to shut off a campaign that produces pipeline while you work on the answer. “Earn first” means make the citation your first new priority, not pull the plug on revenue. If you are still asking whether visibility can simply be purchased, read these seven questions business owners ask before buying AI visibility before you sign anything.

A shopkeeper paying rent beneath a competitor’s glowing answer sign

The earn budget belongs on pages, not just charts

This is where teams can agree with my verdict and still waste the money. They buy a dashboard that watches citations move up and down, then call the chart progress. I have done it. You get a share-of-voice view, a weekly email and the same pages that were not citable last month. Measurement matters; it just is not the work the measurement is supposed to direct.

The market splits into monitors and tools that act. Peec AI Starter at €89 a month and Ahrefs Brand Radar at $199 a month per index are monitoring-only, without optimization or content generation. Profound, at $99 to $399 a month plus enterprise, added Agents that draft content and run gap analysis; Scrunch, at $250 to $500 a month, pivoted to optimization. One category tells you the score. The other can help change what is on the field. Do not pay the second category’s price expecting action and accept only the first category’s report.

A monitor asks, “Where did we get mentioned?” An actor asks, “Which paragraph on which page should the model have cited instead, and what is missing from it?” That second question is the job. The useful work is to rewrite the thin section, add a direct answer, fix the structure a crawler tripped on, publish the page and check whether the citation appeared. A tool cannot promise that last step will go your way. It should at least show what changed before you evaluate the outcome. If all it can name is a ranking or a prompt, you bought a television for watching the problem.

For that $20k client, I gave him an order, not a split. Keep paid running on the five to ten queries that already produce pipeline. Pausing revenue to fund a citation project is how you lose the room. Then put the earn work on those same queries, not on fifty informational posts. Rewrite one relevant page per query so a model can lift a clean answer from it: a direct 40-word answer near the top, proof beneath it and structure a crawler can parse without guessing. You do not need more content. You need fewer pages that answer better.

The vendor filter follows from that work. Ask what page changed, what changed on it and which prompt you will check afterward. I keep a short list of tools that help you get featured in Google AI Overviews rather than only monitor them for this reason. The distinction shows up when one vendor ships pages and the other ships charts. Ask for the action log before the price. groas sits on the execution side by design, with a named strategist attached, while a monitoring dashboard leaves the rewriting to you.

Monday morning version, since a verdict without a task list does not survive contact with an inbox:

  1. Pull last month’s search terms, sort by spend and mark the ten queries that already produce pipeline.
  2. Check each query in an AI Overview and in AI Mode. Where you pay but are not cited, put the relevant page first in the rewrite queue.
  3. Where you are cited but do not buy, put that query first in the paid test.

One list, both sides, the same ten queries. Paid covers the quarter; earned work gives the next quarter a chance to cost less per visit. I would rather own the paragraph and rent the box than rent the box under someone else’s paragraph.