Last quarter, a concrete contractor we work with in the Minneapolis metro picked up 1,210 more search impressions than the quarter before. Two of their strongest local queries earned no clicks at all.
That gap is the whole problem, and it is why the AI SEO workflow below starts somewhere different than most advice you have read. Not with a tool. With your own search data, and with an honest look at which rankings are still worth anything.
Here is the process we run, the step almost everyone skips, and the point where doing it yourself stops making sense.
Why most AI SEO advice sends you shopping instead of working
Search the topic and you get tool roundups. Thirteen tools, five tools, the seven best platforms for 2026. Each one tells you what to buy and none of them tell you what order to do the work in.
The tools are mostly fine. That is not the issue. The issue is that a tool answers “what should I look at,” and the thing you actually lack is a repeatable sequence with decision points in it, so that the work happens on a Tuesday in February whether or not anyone feels inspired.
You can own Semrush and Surfer and still have published nothing in five months.
The 23x statistic you keep seeing does not describe your business
You have probably seen the claim that AI search visitors convert 23 times better than organic search visitors. It comes from Ahrefs, published June 16, 2025, and it is real: AI search made up 0.5% of their traffic over 30 days and produced 12.1% of their signups.
It is also Ahrefs measuring Ahrefs.
They are a software company with a free-tool funnel, a large library of educational content, and an audience of marketers who sign up for products online at two in the morning. In the same post they note their visitors tend to arrive already deep into the research process. That describes their business. It does not describe a manufacturer quoting custom tooling, or a contractor who books work over the phone after a site visit.
When an agency quotes that number at you, ask whose data it is. The honest answer is that nobody has published a reliable AI-search conversion benchmark for home services, manufacturing, or most of ecommerce. We do not have one either.
What the click data actually shows
Fewer clicks, and the loss is concentrated in a specific place.
The Pew Research Center tracked the browsing of 900 US adults across 68,879 Google searches in March 2025. When an AI summary appeared, people clicked a traditional result 8% of the time, against 15% when no summary appeared. Clicks on the links inside the summaries happened on 1% of visits. People also ended their session more often after a page with a summary, 26% versus 16%.
One correction worth making, because this number gets mangled constantly. You will see it claimed that AI features now touch 58% of all searches. Pew’s 58% is the share of people who ran at least one search that month producing an AI summary. At the query level it was 12,593 of 68,879, closer to 18%. Big number either way. Not the number people keep citing.
Now here is what that looks like inside a real account.
Position 7 is the new page 2
In our concrete client’s Search Console data for the last three months, queries ranking in the middle of page one produced almost nothing:
| Query | Avg position | Impressions | Clicks |
| concrete services minneapolis | 9.0 | 312 | 0 |
| concrete contractors st. louis park | 7.8 | 216 | 0 |
| quality concrete (head term) | 7.9 | 1,618 | 16 |
| quality concrete services | 4.3 | 445 | 36 |
Read the last two rows again. The head term sat at position 7.9, pulled 1,618 impressions, and returned 16 clicks. A narrower term sat at position 4.3, pulled less than a third of the impressions, and returned more than twice the clicks.
The standard advice in this business is to hunt for “striking distance” keywords in positions 4 through 10 and nudge them up. That heuristic is now half wrong. Positions 4 through 6 still pay. Positions 7 through 10 increasingly do not, because an AI summary, a map pack, and a few ads sit above them and the user has their answer before they scroll.
Account-wide across that same band, the client gained 43% more impressions and 31% more clicks quarter over quarter while their average position improved from 8.1 to 6.7. So this is not a disaster story. It is a targeting story: the work paid off precisely where it moved things into the top five, and produced nearly nothing everywhere else.
One account, one industry, one metro. Do not treat it as a law. Check your own numbers against it.
The six-step AI SEO workflow
Six steps, unevenly weighted. Steps one and six take the most judgment and the least software.
1. Start with your own Search Console data, not a keyword tool
Open Search Console, set the range to the last three months, and sort your queries by impressions. You are looking for terms sitting between positions 4 and 6 with impressions and weak clicks. Those are reachable. Terms parked at 8 or 9 need a different decision: either commit real effort to move them inside the top five, or let them go.
Keyword tools estimate. Search Console reports what actually happened on your site. Start with the one that is true.
2. Read your query list for sentences, not keywords
This is the step nobody tells you about, and it takes ten minutes.
Scroll your query list for anything written like a sentence. In a metal machinery client’s account we found “where can i buy a new tech machinery rollformer in the united states?” pulling impressions at position 8.8. In Eclipse’s own account: “best us-time-zone b2b commerce agency under $750k for a mid-tier manufacturer.”
Nobody types that into a search box. That is a query with a budget ceiling, a geography, and a buyer profile inside it, which is what a question looks like after it has passed through an AI assistant. When those start showing up in your Search Console, you are already being surfaced in AI-mediated search, and the phrasing tells you exactly what your buyers are trying to figure out.
Write those down. They are your next six months of content topics, in your customers’ words, for free.
3. Cluster with AI, then cut about a third
Paste your query list into ChatGPT or Claude and ask it to group the terms by the underlying job the searcher is trying to do, not by keyword similarity. This part AI is genuinely good at, and it takes two minutes instead of an afternoon in a spreadsheet. It is the same principle behind anywhere else in the business: let the machine do the sorting, keep the judgment for yourself.
Then cut roughly a third of what comes back. The model will confidently cluster queries that are irrelevant to you, seasonal, or already served by a page you forgot you had. You know your business and it does not.
4. Write it yourself, use AI on the structure
Unpopular position, and we will hold it: have AI outline, not draft.
Ask it what questions a piece has to answer, what order they belong in, and what a competitor is likely missing. Then write the prose yourself, or have whoever knows the trade dictate it while someone types.
The cost of doing it the other way is not a Google penalty. It is that AI-drafted copy contains no specifics, because the model does not know your lead times, your minimum order, or why you stopped offering a service in 2023. Specifics are the entire reason someone chooses you, and they are the thing an AI engine can quote.
If you must draft with AI to get moving, budget real editing time. Thirty minutes of a knowledgeable person adding real numbers beats three hours of polishing generic sentences.
5. Make every section answer its own question
AI engines quote passages, not pages. A section that opens with “as we discussed above” cannot be lifted out and cited, because it does not stand up alone.
So: use a question as the heading where the query is a question, answer it in the first sentence beneath, and make each section survive being read in isolation. This is not a trick for robots. It is also how a busy person skims.
6. Fix the source data the AI engines actually read
Your Google Business Profile, your major directory listings, your hours, your service area, your NAP consistency. Boring, and it is where wrong answers come from.
There is no edit button inside ChatGPT. You cannot correct an answer directly. Fixing the underlying sources and waiting for re-crawl is the only lever available, which is why it is worth doing before you spend on content.
How do you know if it’s working?
You check it by hand, monthly, and you accept that the measurement is worse than what you are used to.
Write down five to ten questions a real buyer would ask, phrased the way a person talks. “Best concrete contractor in St. Louis Park.” “Who can build a rollformer for a mid-size shop.” Run them through ChatGPT, Perplexity, and Google, on the same date each month. Record whether you appear, whether you are cited with a link, and who shows up instead.
That is tedious and it works. Once a month is enough to see a trend.
Before you buy an AI visibility platform, know what you are buying. Every vendor computes its own visibility score with its own prompt set, its own engine mix, and its own weighting, so two tools will hand you two different numbers for the same brand and neither is wrong. Treat any score as a trend line on your own fixed prompt set, never as a grade you can compare to a competitor’s.
We use SEOcrawl and Ahrefs for this. Both are useful. Neither will tell you what your number “should” be.
Who should not run this workflow themselves
Plenty of businesses should run it in house, and if that is you, take the six steps above and go.
Here is the honest constraint. This workflow needs roughly six to ten hours a week, sustained across two or three quarters, from someone who can read Search Console without guessing at it. That is our own estimate from client work, not an industry statistic. The failure mode is almost never the tools. It is month four, when the person who owned this got pulled onto a job site and the monthly check stopped happening.
So the real question is not whether you can do this. It is whether anyone on your payroll owns it on a bad week.
If the answer is no, buying more software will not fix it, and you are better off either hiring for the role or bringing in help. That is the work we do, and you can see how we scope it. For reference, our fixed-scope engagements start at $8,000, and most clients move to a retainer afterward because this is maintenance work, not a project you finish. If that number is out of range right now, run the workflow yourself. It genuinely works. It just needs an owner.
Start here this week
Pick five questions your best customer would ask out loud. Run them through ChatGPT and Google, write down what comes back, and note who got named instead of you.
That takes twenty minutes and it will tell you more about where you stand than any audit you could buy. If you would rather hand the whole thing to someone.
Common questions about AI SEO workflows
Ask it the questions your buyers ask, phrased naturally, and read the answer. Use prompts like “best [your service] in [your city]” or “who can [solve my specific problem] near [location].” Note whether you appear, whether the answer links to your site, and which competitors get named. Free tools exist that automate this, but the manual version costs nothing and shows you the actual wording.
Monthly is enough. Pick a consistent date, run the same fixed set of prompts, and log the results in a spreadsheet. Checking more often mostly captures noise, since answers vary between runs even with identical prompts. What matters is the direction over several months, not any single result. Quarterly is acceptable if monthly will not realistically happen.
Not directly. There is no dashboard where you edit your own entry. What you can change is the underlying information the models draw from: your website, your Google Business Profile, directory listings, and third-party mentions. Correct inconsistencies at the source, and answers tend to self-correct over subsequent crawls. Expect weeks, not days.
No. Each vendor builds its own score using a different prompt set, a different mix of AI engines, and its own weighting formula. Two platforms will report different numbers for the same brand on the same day, and both can be internally correct. Use a score as a trend line within one tool over time. Do not compare it against a competitor’s number from a different tool.
No, and treating them as separate budgets causes problems. The same fundamentals drive both: content that answers real questions, clear site structure, accurate business information, and credible third-party mentions. AI engines lean heavily on conventional search infrastructure to find and rank sources. The practical change is in how you structure content and what you measure, not a different discipline.
We have not found reliable published data on AI-search-specific timelines, and treat any agency promising a number with suspicion. From our own client work, expect the first movement in Search Console impressions around month two or three, with click and citation changes following later. Two to three quarters of consistent work is a realistic planning horizon. Anyone guaranteeing faster is guessing.
