Your SEO Didn't Break — Search Did

The short version

If your rankings are stable and your organic traffic is down, you probably have not been penalized. Roughly 68% of US searches now end without any click, and on queries where Google shows an AI Overview, click-through runs at less than half the rate it does without one. Pew found users click a traditional result 8% of the time when an AI Overview is present, versus 15% when it is not.

Ranking third increasingly means being invisible, because the answer was assembled above you.

The response is not a new vendor. It is being the source those answers get built from.

What actually changed

Search used to be a routing layer: you asked, it gave you ten places to look, you clicked one. Now, for a large and growing share of queries, it is an answer layer that reads the ten places for you and summarizes.

The numbers on how much that costs publishers vary by methodology, but they all point the same direction. Ahrefs measured roughly 58% lower CTR on top-ranking pages for AI Overview keywords. Organic CTR on AI Overview queries bottomed near 1.3% in December 2025 and recovered to about 2.4% by February 2026 — still well below the roughly 3.3% on queries with no AI Overview. Google Search traffic to publishers fell by about a third globally in the year to November.

Two implications that follow directly:

1. Rank is necessary but no longer sufficient. You have to be in the answer, not just in the results. 2. The queries you lose first are the informational ones. "How much does X cost" gets answered above the fold. "Book an appointment with X" still clicks through. If your traffic mix skewed informational, your drop is steeper than your competitor's, and it says nothing about your site quality.

What answer engines actually favor

Nobody outside these companies knows the ranking functions, and anyone claiming otherwise is selling something. But the observable behaviour is consistent enough to act on:

• Extractable claims. Short, self-contained, declarative statements with a specific number, range, or timeframe. A sentence that needs the surrounding 400 words to make sense cannot be lifted into an answer.
• Corroboration across independent sources. Being described the same way on your site, your Google Business Profile, a trade association listing, and two review platforms carries more weight than one high-authority backlink.
• Resolvable entity identity. The model has to be confident about who you are before it recommends you. This is where most local businesses quietly fail.
• Specificity that is scarce. Models have infinite generic content and very little concrete data. Real prices, real timelines, real eligibility rules and real limitations are disproportionately citable *because* almost nobody publishes them.

Do this in order

1. Baseline before you change anything

Run 15-20 real buyer questions through ChatGPT, Perplexity, Google AI Mode, and Copilot. Use the language customers actually use, not keywords. For each, record: were you named, who was named instead, and which sources were cited.

That citation list is your real competitive set. It is frequently not the competitor set you assumed, and it frequently includes directories and aggregators rather than businesses.

2. Fix your entity footprint before touching content

Exact-match name, address, and phone — character for character — across your site, Google Business Profile, Apple Business Connect, Bing Places, and the top directories in your vertical.

The killers are small: "Suite 200" versus "Ste 200", "&" versus "and", an old tracking number, an LLC suffix present in some listings and not others. Inconsistency here is the single most common reason a business that should be recommended is not, because the model cannot confidently resolve which listing is you.

This step is boring and it gates everything after it. Do it first.

3. Move the answer to the top of the page

Most SMB pages open with a paragraph about how long you have been family-owned. Rewrite so the first 40 words answer the question the page targets:

> "Emergency drain clearing in Dallas costs $250-$450 for most residential jobs, and we arrive within 90 minutes for calls placed before 8pm."

Then the credibility, the story, and the CTA. The test: could those first 40 words appear verbatim in an AI answer and still be useful? If they need the rest of the page, rewrite them.

4. Publish the specifics your competitors will not

Prices. Timelines. Eligibility rules. What you do not do. Honest guidance on when someone should not hire you.

This is uncomfortable and it is the most durable advantage available, because a competitor can copy your page structure in an afternoon and cannot copy a willingness to publish real numbers. Date-stamp anything that will go stale — a cited 2026 price still circulating in 2028 is a problem you created.

5. Structured data and machine-readable surfaces

LocalBusiness, FAQPage, and Service schema with real questions and real answers. An honest sitemap. If you publish substantial reference content, an `llms.txt` index helps AI crawlers understand the shape of the site. None of this is magic — it removes ambiguity so extraction is cheap.

6. Re-run the baseline monthly

Same questions, same engines. Track one number: in how many answers were you named. Expect this to move slower and noisier than rankings did. Judge it on a two-quarter trend, not month to month.

Be sceptical of the vendors arriving right now

"AEO" and "GEO" retainers are being sold hard, frequently with implied placement guarantees.

Nobody controls what a model cites. The mechanisms are not fully observable, the outputs vary between users and sessions, and there is no submission endpoint. Any guarantee of placement in an AI answer is either a misunderstanding or a lie, and it is a disqualifying signal either way.

The honest version of this work is content and entity hygiene. It is unglamorous, it is mostly a few days of effort rather than a platform purchase, and it is the same work that made you rank in the first place — pointed at a reader that is now a model instead of a person.

The floor under all of this

The businesses least affected by any of this are the ones whose demand does not route through search at all: referral engines, repeat customers, and owned email lists. If search was your only channel, the structural fix is not better AEO — it is a second channel.

The full implementation, with a baseline audit sheet, the answer-first rewrite pattern, and an entity consistency checklist, is written up as a workflow: Answer Engine Visibility. For the owned-channel side, start with reactivating cold leads and review generation — both compound in a way rankings no longer do.