Every article about AI and SEO is written by someone selling an AI SEO tool. This one is written from our own Search Console, and the numbers are not flattering to us.
We rank on page one for a commercial keyword. It sends us almost nothing. Here is exactly what that looks like, why it happens, how to check the same thing on your own sites in about ten minutes, and what we think agencies should do instead of buying another subscription.
The short answer
| Question | Answer |
|---|---|
| Has AI changed SEO? | It has broken the link between ranking and traffic that the reporting model was built on. Ranking still matters, but it no longer converts into visits at historical rates. |
| What is our evidence? | One of our pages ranks at an average position of 9.3 and gets a 0.49% click-through rate, from 16,064 impressions and 79 clicks. |
| What is the fingerprint of AI-mediated search? | A large share of impressions from queries too rare for Google to name. For that page, 62.2% of impressions sit in that anonymised bucket. |
| Should we stop doing SEO? | No. Change what you optimise for and what you report. |
| What should we write? | Content that can be quoted: direct answers, comparison tables, named criteria and first-party numbers. |
| Which metric should replace position as the headline? | Branded search, direct traffic, assisted conversions and, where possible, citation appearance in assistants. |
| Do we need an AI SEO tool? | Only when you can name the decision it will change. Search Console already tells you the most important thing for free. |
The rule that matters most: stop reporting average position to clients as a proxy for traffic. Report what the business actually received: enquiries, branded searches and conversions.
The AI SEO number that should change how you report to clients
One of our pages, a comparison article targeting a commercial buying keyword:
| Metric | Value |
|---|---|
| Impressions | 16,064 |
| Clicks | 79 |
| Click-through rate | 0.49% |
| Average position | 9.3 |
Period: 11 June to 7 September 2026, 89 days. Source: Google Search Console, sc-domain:inflowave.io, filtered to that single URL. Pulled 9 September 2026.
Position 9.3 has historically returned somewhere between 1.5% and 3% click-through. We are getting 0.49%. The page is not broken. It is indexed, it renders server-side, it carries Article, ItemList, FAQPage and SoftwareApplication schema, and its title is fine. It ranks. It simply does not get clicked.
To put the gap in plain numbers, a 1.5% to 3% click-through on 16,064 impressions would be roughly 240 to 480 clicks. We got 79.
Historical click-through at a position near 9.3 against what our page received. Same position, far fewer visits.
If you are still reporting average position to clients as a proxy for traffic, that relationship has quietly stopped holding.
What our Search Console query log shows about AI SEO
Search Console lists 141 individual queries for that page. The interesting ones are not keywords at all.
Real queries, verbatim, with their position and click count:
- "best platform to manage whatsapp and instagram dms with a shared agent inbox" - position 1, zero clicks
- "what is the best software for managing customer conversations across whatsapp, instagram, and facebook in one inbox?" - position 1, zero clicks
- "our managers need a dashboard of first-response times, agent workload and sales conversion across whatsapp, instagram and facebook. which chat platform integrates with our crm to show these metrics in one place?" - position 1, zero clicks
- "best omnichannel crm for businesses managing sales over whatsapp instagram and email" - 193 impressions, position 3.2, zero clicks
- "what is the best crm for tracking instagram handles?" - 190 impressions, position 2.7, zero clicks
Position one, zero clicks, repeatedly.
One "query" is a 500-word buyer persona document that somebody pasted into a search box. It lists key personas, business context, buying triggers, decision criteria, organisational change tolerance and an AI boundary section, and ends with the actual question. That is not a search. That is a prompt.
62.2% of that page's impressions come from queries too rare for Google to name. 6,070 of 16,064 impressions are attributable to the 141 listed queries. The other 9,994 sit in the anonymised bucket, which is where unique, never-repeated, sentence-length queries end up.
Where the impressions came from. The larger bar is the anonymised bucket, the part of Search Console you cannot see query by query.
That distribution is the fingerprint of AI-mediated search. People are not typing "instagram crm" any more. They are describing their situation in full paragraphs and reading a synthesised answer.
How to check your own anonymised-query ratio
You can run the same check on any client site in Search Console in about ten minutes. It costs nothing.
- Open the Performance report for the property and set the date range to the last three months.
- Add a Page filter for one commercial URL, such as a comparison or service page.
- Note the total impressions shown at the top of the report.
- Open the Queries tab and add up the impressions of the queries listed. Search Console omits anonymised queries from that table, so the listed rows add up to less than the total.
- Subtract the listed total from the overall total. The difference is the anonymised share. In our case, 16,064 minus 6,070 leaves 9,994, or 62.2%.
- Repeat for your top five commercial pages and note which ones have a rising share over time.
| What you see | What it suggests | What to do |
|---|---|---|
| Low anonymised share, normal click-through for the position | Searchers are still typing short keywords and clicking | Keep doing what works, and keep reporting it |
| High anonymised share, low click-through | Visibility has moved into long, conversational questions and AI answers | Rewrite the page to be extractable, and report conversions instead of position |
| High impressions growth, flat clicks | Impressions are inflating without visits | Stop leading the client report with impressions |
| Page one for definitions, few clicks | Assistants answer the definition inline | Move effort to pages that need current or first-party data |
What this means commercially
Being cited in an AI answer is not the same as being visited. Our page is doing the first job well and the second job barely at all.
Three consequences agencies should plan around:
Impressions inflate while clicks stay flat. Across the same period, one of our tool pages grew impressions 140% with average position unchanged and click-through falling from 10.6% to 6.5%. If your monthly report leads with impression growth, you can show a chart going up while the business gets nothing. That is worse than no report.
Informational keywords are the worst affected. Definitional and how-to queries are exactly what an assistant answers inline. If a content plan is full of "what is" and "how to" topics, it is optimised for the queries least likely to produce a visit.
"I have had B2B SaaS clients building endless blogs around informational keywords come to me with 75% traffic losses and no position shifts," says Max Desiak, a B2B SaaS SEO & AEO consultant. "What is interesting is that almost none of them saw any serious negative impact on demos or pipeline. So it is not as bad as it looks."
Your best content may now be a sales asset rather than a traffic asset. A thorough comparison page that gets quoted to a buyer mid-research still influences the decision. It just will not show up as a session.
What actually works in AI SEO now
None of this is a reason to stop. It is a reason to change what you optimise for.
Write to be quoted, not just ranked. Assistants extract. Direct answers near the top, explicit comparison tables, named criteria, real prices, clear definitional sentences. Burying the answer under 600 words of preamble is now actively costly.
Target the queries assistants cannot resolve alone. Anything needing current pricing, a real comparison across named products, or first-party data. Generic explainers are commodity output. Original measurement is not, which is the entire reason this article exists.
Publish numbers only you have. We can publish our Search Console data. A tool vendor writing about click-through decline cannot publish yours. First-party evidence is the one axis where a small site beats a large one, because authority does not substitute for access.
Stop treating position as the KPI. Track branded search volume, direct traffic, assisted conversions and, where you can, citation appearance in assistants. Position alone now overstates performance.
Keep Search Console central. It is free, it reports real impressions and clicks including AI-shaped surfaces, and the anonymised-query ratio is a usable proxy for how much of your visibility has moved into AI answers.
How to write a page that can be quoted
A practical layout for a commercial page, applied to any client:
- Open with the direct answer in the first two or three sentences, before any background.
- Add a short table of key questions and answers straight after, so the answer can be lifted cleanly.
- Name your criteria. If you compare options, say what you compared them on and why.
- Use real figures and say where they came from, with the date you checked them.
- Add a comparison table for the choices a buyer is weighing.
- Finish with a question-and-answer section phrased the way people actually ask.
- Say what you could not verify. Honest gaps are more quotable than confident guesses.
The shape of a page built to be quoted. The first two steps are the ones an assistant is most likely to lift.
Old SEO reporting against reporting that holds up
Same data source, different headline. The right-hand report is the one a client can connect to money.
| Metric | What it tells you now | How much to trust it alone |
|---|---|---|
| Average position | Where the page ranks | Low. It overstates performance when clicks have fallen |
| Impressions | How often the page was shown | Low. It can rise while visits fall |
| Click-through rate | Whether being shown turns into a visit | Medium. Compare with the usual rate for the position |
| Branded search | Whether people are looking for the business by name | High. A sign that content is building awareness |
| Direct traffic | Whether people return without searching | Medium to high |
| Assisted conversions | Whether content helped before the enquiry | High, if you track the whole journey |
| Enquiries and revenue | What the client actually wanted | Highest |
AI SEO tools for agencies, honestly
Prices below were read from each vendor's own pricing page on 9 September 2026. Verify before quoting to a client, because this category changes pricing constantly.
AI visibility tracking
Profound. $99/month and $399/month self-serve tiers plus custom enterprise. Widely described in roundups as enterprise-only, which is no longer true. Prompt-level tracking and historical data. The strongest option if you need to show clients where they appear in assistant answers.
Peec AI and Otterly.ai are the mid-market and entry-level alternatives most commonly named. We could not verify either price at source (Peec renders pricing in JavaScript, Otterly returns a bot challenge), so we are not quoting figures other articles have copied from each other.
Content and optimisation
Surfer SEO. Discovery EUR 49, Analytics EUR 82, Standard EUR 99, Pro EUR 182, Peace of Mind EUR 299, Enterprise EUR 999 per month billed yearly. The plan names have been rebuilt around AI visibility rather than classic content scoring.
Frase. $39/month billed yearly ($49 monthly) for solo use, $103/month ($129 monthly) for teams including 40 articles and 250 audit pages, and a $299/month tier. Still one of the better value briefs-and-optimisation tools.
The free ones that matter more than the paid ones
Google Search Console. Everything in this article came from it. Nothing you buy will tell you more about your own AI-era performance than the anonymised-query ratio you already have for free.
Screaming Frog. Still the fastest way to find the technical problems that no amount of AI content will outrun.
A category to approach carefully
Several AI SEO platforms bundle a backlink network as a headline feature, offering contextual links from thousands of partner sites. Those partner sites are usually other customers of the same platform. Reciprocal link networks operated by a single vendor are a link scheme under Google's spam policies, and the shared platform is exactly what makes the footprint detectable.
For an agency the asymmetry is bad. The upside is faster link acquisition. The downside is a manual action on a client's domain, on a site you do not own, for a subscription you chose. Read what the link feature actually is before putting client sites into one.
Mistakes agencies make with AI SEO
- Leading the report with impressions. The chart goes up and the business gets nothing.
- Filling the content plan with "what is" pages. These are the queries assistants answer inline.
- Buying a tracker with no plan for the result. A dashboard nobody acts on is a cost.
- Copying prices from roundups. Prices in this category change, and many figures are copied between articles. Read the vendor's own page.
- Joining a link network because it is bundled. A vendor-run network of customer sites is a risk to the client's domain.
- Publishing generic explainers at scale. Volume with no editorial judgement is the failure mode, not the use of AI itself.
- Not telling the client early. Change the client report before the client changes it for you.
What we would actually do
If you run an agency and you are deciding where to put effort this quarter:
- Pull the anonymised-query ratio for your top five commercial pages. That tells you how much of your visibility has already moved into AI answers. It costs nothing.
- Change the client report before the client changes it for you. Lead with branded search, direct traffic and conversions. Keep position as context, not as the headline.
- Rewrite your best commercial pages to be extractable. Answer first, table second, prose third.
- Pick one thing you can measure that nobody else can, and publish it. This is the highest-leverage content you can produce right now.
- Buy a visibility tracker only once you can name the decision it will change.
If the reporting problem above is really a systems problem, the tool that holds the client relationship matters more than the tool that measures the ranking. We compared the options in the best CRM for SEO agencies.
The honest summary: AI has not killed SEO, but it has broken the link between ranking and traffic that the entire reporting model was built on. The agencies that adjust what they measure will keep the retainer. The ones still selling position will lose it the first time a client checks their analytics.
Where Inflowave fits (and where it does not)
Inflowave is a CRM for agencies, not an SEO rank tracker or an AI visibility tool, so it does not measure where your pages appear in assistants. What it can do is hold the part that counts once a visit happens: the lead record, where the lead came from, and what happened to it afterwards. Its analytics include a sales-by-source view that splits inbound from cold, which is the kind of number that lets you report enquiries and revenue rather than position. See Inflowave analytics, and tracking links for attributing a visit to a specific page or campaign.
If you are building the client reporting side of this, client reporting software for agencies covers what a monthly report should contain. Position is context. The business result is the headline.
Questions people ask
Does ranking still matter for SEO?
Yes, but as an input rather than an outcome. You generally have to rank to be cited. You just cannot assume that ranking converts into a visit at historical rates. Our own page ranks 9.3 and converts 0.49% of impressions into clicks.
Is AI-generated content penalised by Google?
Google's stated position is that it rewards helpful content regardless of production method, and penalises scaled content abuse. In practice the risk is not the generation, it is publishing volume with no editorial judgement, which is also the failure mode of most autoblogging tools.
Should agencies buy an AI visibility tracker?
Only when you can name the decision it will change. If it produces a slide nobody acts on, it is a reporting cost. If it tells you which pages to rewrite for citation, it earns its price.
What is the single most useful free signal?
The share of your impressions that Google cannot attribute to a named query. It is in Search Console today, and a rising ratio means your audience has moved into assistants.
Why do I have impressions but no clicks?
Your page is being shown but the searcher gets what they need without visiting, or the result does not look like the answer. In our data, queries at position one produced zero clicks. Check the click-through rate against what is normal for your position, then look at whether an assistant answers the question inline.
What are anonymised queries in Search Console?
They are searches that Google leaves out of the query table, typically because they are rare or unique. The listed queries add up to less than the total, and the difference is the anonymised share. For our page that share was 62.2% of impressions.
What should I report to clients instead of average position?
Lead with branded search, direct traffic, assisted conversions and enquiries or revenue. Keep position and impressions as context. A client cares about business results, and position alone now overstates performance.
Is it worth writing "what is" and "how to" content?
Less than it used to be for traffic, because assistants answer those questions inline. It can still be worth it as a sales asset, as supporting content or when you add first-party data an assistant cannot produce. Do not fill the whole plan with it.
Are AI SEO backlink networks safe for client sites?
Treat them with great care. When a vendor's links come from other customers of the same platform, that is a reciprocal network operated by one vendor, which is a link scheme under Google's spam policies. The downside lands on the client's domain, so read what the feature actually is first.
How do I make a page more likely to be cited by an AI assistant?
Put the direct answer first, follow it with a table, name your criteria, use real figures with their source and date, and end with questions phrased as people ask them. Assistants extract, so a page that is easy to lift from is easier to cite. Citations are not guaranteed, so measure what you can.
