DEV Community: reidify

Search Console now counts searches made with a camera. It cannot tell you what was photographed.

reidify — Fri, 02 Oct 2026 06:27:34 +0000

For the first time, Search Console reports the searches people make by pointing a camera at something. What it leaves out matters as much as what it adds: there is no query.

On 24 September, Google announced web multimodal reporting in two places: the Performance report for Search results, and the report for Generative AI features. Google says the data covers four routes: Lens, Circle to Search on Android, image uploads to Google Search, and Chrome's right-click "Search this image". The rollout is global and started that day, so some properties will see the filter before others.

The Performance report help page now splits Web in two. "Text-based" is traffic from text typed into the standard search bar. "Multimodal" is traffic from searches made with images.

The query column is gone

This is the part that changes how you read the report. Google's dimensions documentation says it plainly: "Because multimodal searches mostly use images rather than text, specific text query data isn't available for this traffic. As a result, the queries dimension isn't available when this search type is selected."

Most Search Console habits start from the query: which words found you, and which words should have found you. With multimodal traffic that starting point does not exist. You get pages, countries, devices and dates. You do not get what the person was looking at.

So for this traffic, the page is the query. If a product page appears for multimodal searches, someone photographed something that Google decided your page answers. The page list is the only record of intent you have.

There is no before

Google's John Mueller said on Bluesky, as PPC Land reported, that "the data wasn't previously in the counts, the report has new data." Taking that at its word, there is no baseline to compare against. A line that climbs through October may only be the rollout reaching your property. Treat the first weeks as a census, not a trend.

Google's post also names exactly four routes. It does not say where Search Live sessions or images asked inside AI Mode are counted. Until it does, assume the multimodal figure is a floor.

How to read it this week

In the Performance report, set the search type to Web: multimodal. If the filter is missing, either the rollout has not reached your property or Google has no multimodal traffic to report for it yet.
Switch to the Pages tab and export it. That list is your multimodal query list.
For each page, write down what someone would have had to photograph to land there. Pages you expected and do not see are as informative as the ones you do.
Apply the same filter in the Generative AI report. That is where image-led searches meet AI answers.
Before comparing any Web figure across 24 September, check which search type you are looking at. If the multimodal data is new, a date range that spans the rollout is not like for like.

Google has not published how an image gets matched to a page, and we would not trust anyone who claims to know the weighting. The report gives you the outcome, not the mechanism. When the input is a picture, a page has to make clear what it is about to anything reading it: what the thing is, what it is called, and what it does. That is the same discipline behind an AI-readable website, arriving through a different door.

The camera has always been a search box. Now it shows up in the report, and the page is where you read what people were asking.

Every link in an AI Overview gets the same position in Search Console. Google confirmed it.

reidify — Fri, 25 Sep 2026 14:41:19 +0000

Search Console's AI Overview numbers do not measure what they appear to measure, and Google said so in public this month.

On 13 September, Search Engine Journal reported a Reddit thread in which Google's John Mueller agreed with a site owner's breakdown of how the generative AI performance report counts. Search Engine Watch covered it the next day. The rules themselves are written down in Google's own help page on impressions, position and clicks. What changed is Google saying out loud that they do not produce a useful answer to the question most people open the report to ask.

Mueller's words: "Position for these is hard to do in a way that makes it useful."

The position belongs to the box

Google's documentation treats an AI Overview as one result occupying one position. Every link inside it is assigned that same position.

So if an AI Overview cites five sources, the first and the fifth report identical positions. A better average position can mean the AI Overview moved up the page. It cannot tell you that your citation moved up inside it. The number describes the container, not your place in it.

Impressions miss in both directions

AI Overviews follow the standard impression rules. As the thread Mueller endorsed put it, an impression counts when the result is on the page that was served, whether or not anyone scrolled to it. That inflates exposure.

Links hidden behind the expansion are the opposite case. They are not counted until someone expands the answer. That understates exposure.

The chart shows one line. It does not show which of the two errors dominates for your pages.

Do not add it to your Web totals

The AI report is a filtered view of data already inside the Web search type, not a separate layer on top of it. Adding the two double counts. Dividing all Web clicks by AI impressions produces a number that looks like an AI click-through rate and is not one. The AI report carries no click data of its own.

What the report is still good for

It answers presence questions well: which of your pages appear in AI Overviews and AI Mode, in which countries, and how that changes over time. It does not answer prominence questions: how visible your citation was, or whether it earned the visit.

When the report finished rolling out, we argued that its first readings should be written down with a date rather than treated as a verdict. This is the second half of that argument. Even a clean reading is a reading of presence.

How to read it

Read position as the position of the AI Overview, never as yours inside it.
Read an impression as "was on the page", not "was seen".
Keep the AI report out of any sum with Web totals.
For prominence, look at the answer itself. Run the query, read what comes back, and record where you are cited and in what order. That is the manual half of measuring AI visibility, and it is still the only way to see order inside the answer.

Mueller invited suggestions for a better position metric. Neither report mentions a replacement or a date. Until one ships, the report tells you that you were in the answer. It does not tell you how well you were placed.

Google now lets any site leave AI Overviews. Most businesses should not.

reidify — Fri, 18 Sep 2026 03:51:07 +0000

Every website in the world now has a switch that removes it from Google's AI answers. Google added a note to its documentation confirming it: "As of August 31, 2026, we've rolled out this control to all websites worldwide." (Search Engine Land)

The switch is real, it is free, and for most businesses reading this it is the wrong lever to pull. Here is what it does, and why the people it was built for are not most of you.

What the control actually does

It lives in Search Console, under Settings, and it works at the property level. Turn it on and your site is excluded from AI Overviews, AI Mode and AI Overviews in Discover, both as a visible link and as grounding, meaning the model can no longer draw on your pages to compose an answer. (Relevant Audience)

Google's own framing has three parts worth reading closely:

Sites that opt out "will not receive traffic or impressions" from those features.
The setting is not used as a ranking signal for ordinary search results.
It does nothing to AI training. That is a separate control, Google-Extended, and turning one does not touch the other.

It is also all or nothing. The UK's Competition and Markets Authority has given Google until March 2027 to introduce page level controls, so today you cannot keep your product pages in and your archive out. It is the whole site or none of it.

Who it was built for

The control exists because of regulators, not because site owners asked nicely. Google made the concession in June, on the same day the CMA instructed it to give publishers a way to stop their content feeding its AI features. The European Commission has since sent publishers a questionnaire, reported at the start of September, asking whether they intend to use it and what is shaping that decision, because Brussels wants to know whether leaving costs a publisher its ordinary rankings. (Resultsense, Android Headlines)

That is a publisher's problem. For a news site, the article is the product. An AI summary of it can replace the visit, and the visit is where the ad revenue lived. Opting out is a defensible trade for them.

Why the same trade runs backwards for most businesses

For a clinic, a law firm, a B2B supplier or a studio, the page is not the product. The page exists so that someone, human or machine, picks you. When an AI answer names your firm in response to a buyer's question, that is not traffic stolen from you. That is the outcome the page was written to produce.

Opt out and you lose that outcome, and you lose the only Google instrument that shows whether it is happening. The new generative AI report in Search Console shows impressions from exactly these surfaces. Opted out, there is nothing left for it to count. The report already carries no click data, so you would be trading a partial view for none.

Our own page on how to appear in Google AI Mode argues that broad, clear topic coverage is what gets a site into these answers. This control is the mirror image of that work: one checkbox that undoes all of it, for the whole property, until page level controls arrive.

What to do this week

Open Search Console and confirm the setting is off. It is a new control, and a decision this size should be made on purpose, not discovered later.

If you run an ad-supported publication, it is a genuine decision and the EU questionnaire is asking the right questions. If your website exists to get you hired, bought or booked, leave it off. Being in the answer is the job.

A newsroom just restructured its org chart around losing traffic to AI. Read the memo.

reidify — Fri, 11 Sep 2026 03:49:00 +0000

Gannett, the parent company of USA Today, cut staff and reorganized its audience team this month. The internal memo explaining the move did not point at a soft ad market or a platform algorithm tweak. It named the mechanism directly: "search traffic is under pressure, and platforms are increasingly keeping user experiences to themselves instead of sending them back to us." Search Engine Land reported the memo on September 4, and Nieman Lab corroborated it the same week.

That sentence is worth sitting with for a moment. "AI Overviews might be reducing click-through" has been a hypothesis for over a year, argued from aggregate data that a marketer could cite and a business owner could reasonably discount as noise. This is not that. A national media company with a dedicated SEO desk and a multi-year runway just reorganized its staff around the claim, in writing, in a memo one of its own people leaked.

The part that applies past media

It is tempting to read this as a publisher-specific story: ad-supported newsrooms have always been exposed to referral traffic in a way most businesses are not. But the mechanism the memo names, platforms keeping the user experience to themselves instead of sending a click back, does not care what kind of site sits on the other end. An AI Overview or an AI Mode answer that resolves a query fully is exactly as complete whether the underlying source was a newsroom or a founder's services page. We asked directly whether AI Overviews actually reduce clicks a while back, working from the same category of aggregate evidence Gannett had before this memo. An org chart is a stronger answer than an aggregate study, because nobody restructures a team over a trend they can wait out.

What is actually different here

Two things separate this from the usual "AI is coming for search" news cycle. First, it is a named company, not an anonymous survey respondent. Second, it is a resourcing decision, not a strategy memo. Gannett has run large-scale SEO operations for over a decade and has more data on its own referral patterns than almost anyone. If that team is the one being reorganized around this pressure, the pressure is not speculative.

For a founder-led business, the honest read is that Gannett had the budget, the traffic history, and the technical staff to notice this early and still ended up reorganizing under it. Most small business and independent sites have none of that instrumentation. The traffic drop shows up later and gets attributed to something else, a slow season, a Google algorithm update, a competitor. By the time it is named correctly, a large publisher will already have adjusted its whole operating model around the same cause.

None of this says AI answer visibility is solved by any particular technique, and this piece is not claiming one. It says the size of the company that just restructured around the problem is the actual news. A hypothesis that a hobbyist blog worries about is one thing. A hypothesis a newsroom lays off staff over is a different category of signal entirely.

Sources: Search Engine Land, September 4, 2026; Nieman Lab.

You can now buy ads inside ChatGPT. You still cannot buy a citation.

reidify — Sun, 06 Sep 2026 04:59:00 +0000

On 31 August OpenAI said its advertising business had reached a $1 billion annualized revenue run rate in under 200 days, and that self-service buying through Ads Manager was opening across India, Europe, the Middle East and North Africa. Ads Manager had already been open to small and midsize businesses since May, and OpenAI says those advertisers now represent a material share of the business.

For a small business in Mumbai or Madrid, something concrete changed this week. ChatGPT went from a place you hope gets you recommended to a place you can put a card down and buy a slot.

The paragraph most of the coverage skipped is the one worth reading. In the same announcement, OpenAI states its ads principles plainly: ads in ChatGPT are always clearly labeled and separate from ChatGPT's answers, and advertising does not influence the answers ChatGPT provides.

Take that at face value and it describes two separate systems sharing one screen.

One is bought. You set a budget, you bid, you get a labeled placement. OpenAI reports that CPC and outcome-optimized bidding now account for most campaigns, with a pixel and a conversions API underneath. It behaves like every ad platform you have already used.

The other is retrieved. When someone asks ChatGPT to help them pick an accountant, a supplier, a school, the names inside that answer come from what the model can find and trust about the business. No budget line touches it.

That distinction is not academic, because the failure mode is obvious and it is expensive. A founder buys ChatGPT ads, sees traffic arrive, and concludes the AI visibility problem is handled. It is not handled. The ad ran. The answer sitting above it may still not know the business exists.

The two also fail differently. Stop paying and the ad stops the same day. That is the deal, and it is a fair one for anything time-bound: a launch, a season, a market you are testing. A citation is not rented, so it does not switch off on a billing date, but it also cannot be turned on by Friday. We wrote up how those two trades compare at reidify.design/compare/aeo-vs-ppc.

One caveat, stated rather than buried. The separation between ads and answers is OpenAI's own description of its own product, published while it was announcing growth in that product. It is a stated policy, not an audited finding, and no outside party can currently verify it. Treat it as the operating assumption it is and check your own results against it.

None of this is an argument against buying. Ads inside ChatGPT are a real channel and, in India, a new one as of this week. It is an argument against reading an ad receipt as evidence of anything else.

The more useful question is the one almost nobody has run. Open ChatGPT, ask it the question a customer would actually ask, the one that names the category and the city rather than the company, and see whose names come back. If yours is not there, no amount of ad spend will put it there, because by OpenAI's own account those are two different systems.

Sources:

OpenAI, A milestone in expanding access to AI, 31 August 2026
CNBC, OpenAI's ad business hits $1 billion annualized revenue run rate, 31 August 2026

Google made its own result links unreadable. Rank tracking now costs up to 1,000 requests per query.

reidify — Fri, 04 Sep 2026 04:00:13 +0000

Google confirmed on 26 August 2026 that it is rolling out google.com/goto passthrough links inside its search results. The anchor behind an organic listing used to carry the destination address. It now carries an encoded wrapper that issues a 302 to the real page.

A Google spokesperson told Search Engine Roundtable: "We have a long history of deploying technical measures against evolving forms of abuse, and we regularly take steps to protect our services and users." The statement names abuse. It does not name scraping, rank trackers, or AI companies, and it arrived with no timetable and no technical documentation.

This is no longer a test. Derek Perkins of the rank tracking firm Nozzle reported near total coverage across several residential IP providers on the same day, after roughly four months of intermittent testing. Alex Greenland logged the first public sighting on 23 June. Brodie Clark documented it across both paid and organic listings on 2 July.

The cost is the story

For most of Google's history, a tool could read a results page and parse each destination address out of it without ever contacting the destination site. That assumption sits underneath rank trackers, share of voice dashboards, competitive intelligence platforms, and a long tail of scripts that marketing teams run in house.

The wrapper removes it. Perkins says the links cannot be decoded locally, so a tool has to follow every redirect to learn where it goes. Two obvious workarounds died in the same thread. HEAD requests, which would retrieve the redirect target cheaply without the body, are not allowed through. A GET that stops at the redirect does work, but volume is the constraint: resolving all the links for one five page ranking takes between 500 and 1,000 requests. Rate limiting becomes the binding limit, rather than bandwidth or storage.

One question decides how bad this gets, and nobody answered it. Dominic Woodman asked whether the redirect is static per page. If it is, a large operator builds a lookup table once and reuses it. If it is not, this is a permanent per query tax on every tool that reads a results page.

The second problem is worse than the cost

Joe Manna asked whether resolving a goto link registers as a click. Perkins did not know, and said it would be interesting to see whether Google correlates those events with client side analytics.

Sit with that for a second. If resolution registers server side, the tool measuring your visibility becomes a source of the traffic it is measuring. And the same wrapper doubles as a detection surface for automated clients, separate from its job as a link.

What this does not touch

The manual test. Asking ChatGPT, Perplexity, or Google the question a real customer would type, then reading what comes back, never depended on parsing a results page. We published the five question version of that test well before any of this, and the argument there was that a paid monitoring tool automates the same test rather than replacing it.

That was a claim about signal quality. It has quietly become a claim about whether the automated version still runs reliably at all.

None of this makes rank tracking useless. Vendors adapt, and they always have. But if a dashboard is the only evidence you hold that your site is being found, it is worth knowing three things: the pipe feeding it just narrowed, nobody has said how permanently, and your vendor cannot tell you either yet.

Sources: Search Engine Roundtable and PPC Land.

Google now reports your AI Overviews impressions. Its first drop was a bug.

reidify — Tue, 01 Sep 2026 04:24:50 +0000

As of 31 August 2026, Google's generative AI performance reports are live on all websites worldwide. That is the report showing impressions from AI Overviews, AI Mode and generative AI features in Discover, broken out by page, country and date, with device data in the Search reports. Google announced the completion here: Search Engine Journal, and Search Engine Land covers the same rollout alongside a new control that lets a site opt out of those surfaces without affecting standard rankings.

It does not include click data.

That is the real change and the real limit in one line. Until now, checking whether your pages surfaced inside an AI answer meant opening a chatbot, asking, and writing down what came back, which is still the method for auditing AI visibility yourself across engines Google does not report on. Now one family of surfaces reports on itself.

Before anyone builds a habit on it, there is a second thing from two weeks earlier.

The first drop in this report was not a drop

Data starting 13 August 2026 showed a fall in impressions in the generative AI report. It had not happened. Google confirmed a logging issue, said it did not represent a change in Search visibility, and said it would add an annotation in Search Console. Search Engine Land reported it, Search Engine Roundtable had the original thread, and it was resolved a few days after being raised.

So a lot of site owners are about to open a brand new chart for the first time, scroll back through the available history, and find a decline that never occurred.

Three things that follow

Do not act on a single fall. A new report is an instrument that is still shipping. Check the dates of a dip against known issues before you rewrite a page, and wait for the annotation Google promised rather than treating the line as settled.

Impressions are not traffic. The report answers whether you were shown, not whether anyone arrived. Those two questions came apart the moment AI answers started resolving the query on the results page, and this report currently measures only the first one.

It covers Google. AI Overviews, AI Mode and Discover. ChatGPT, Perplexity and Copilot are not in it and still have to be checked by hand, one prompt at a time.

Two more limits are worth knowing before you go looking. Google's own help pages note that not every property has access yet, and that a site without enough AI impressions may see no report at all. An empty report is not proof of invisibility. It can just mean the threshold was not met.

What to do this week

Open the report. Write down the number and the date next to it.

That is the whole task. The value is not this month's figure, which you cannot compare to anything. It is having twelve readings a year from now, taken from an instrument whose failure modes you already understand, rather than one number you found in a panic and could not put in context.

A measurement that arrives with a known bug in its first month is not a bad measurement. It is a young one. The mistake would be to treat the first chart it draws as a verdict on work you did months ago.

Reddit lost 86% of its ChatGPT citations in four days, and nobody announced it

reidify — Fri, 28 Aug 2026 05:06:47 +0000

Between 18 July and 7 August, reddit.com held a steady 3.83% of ChatGPT Search citations. Between 14 and 17 August it averaged 0.52%. That is an 86.4% relative decline in four days.

No announcement, no policy post, no deprecation notice. Search Engine Journal's coverage is built around the fact that the drop still is not fully explained, and that is the most useful part of the story.

The obvious suspect does not quite hold. ChatGPT Search changed its query fan-out behaviour on 8 August, and its use of the site: operator jumped from around 0.4% to nearly 17% of fan-out queries that day, then kept climbing. But the main decline started on 14 August. The timing is close enough to be suggestive and far enough apart that nobody can claim certainty.

It was also specific to one engine. Promptwatch found no comparable one-day fall in Google's AI products. Reddit's share of citations in Google AI Overviews drifted from about 2.5% in early July to roughly 2.1% in August. That is a slope, not a cliff.

Sources: Forbes, corroborated by Search Engine Land and Search Engine Journal.

What this means if you own a website

An AI visibility number is a reading, not a rank.

Search positions move slowly, and usually for reasons you can reconstruct afterwards. Citation share in an answer engine can lose most of its value in under a week, from a retrieval change that was never published, on one engine, while the others carry on unchanged. Reddit is among the most cited domains on the internet. It did not get advance notice either.

So the number only means something with its conditions attached: the date, the engine, the questions asked, and the denominator.

We publish our own method at reidify.design/research/measuring-ai-visibility, including the run where we scored ourselves 78% and the corrected run that scored 25%. That page argues quarterly testing is enough for a small business, and this episode does not change it. Chasing a weekly citation number is not a business activity.

What it does change is how much weight one reading can carry. It is a reading of a system that moves on its own, so it belongs in a record with a date on it rather than on a slide as a score.

The failure mode to avoid is the number that gets sold. "Your AI visibility is 62%" invites a founder to hear a property of their website. It is not. It is a property of their website and one engine's retrieval behaviour on one day, and only the first of those is theirs to fix.

If you measure, keep the dates and keep the misses. The misses are what tell you whether the number moved because you did something, or because someone else did.

Four engines, four different answers. We have the receipts for both scales.

reidify — Wed, 26 Aug 2026 16:17:55 +0000

Yext published an analysis of 17.2 million AI citations this month, and the headline is not that AI search is hard. It is that AI search is not one thing.

The four major engines apply fundamentally different retrieval logic. Not different weightings of a shared ranking method. Different logic. Verified, structured, directly distributed data made up 54.53% of distinct citation sources overall, but the mix shifts hard depending on which engine you ask. ChatGPT retrieves largely through Bing's index via OAI-SearchBot and cites seven to eight sources per response. Claude cites user-generated content at two to four times the rate of the other three.

Source: https://www.yext.com/research/ai-citation-behavior-across-models, corroborated by https://cmotech.uk/story/study-finds-ai-models-develop-distinct-citation-habits

That is a large-scale finding, and large-scale findings are easy to nod at and hard to feel. So here is the same thing at a scale of thirty-six.

On 21 August we asked three engines twelve questions about who does answer engine optimisation, in fresh incognito sessions, never naming our own studio. Twelve blind questions, three engines, thirty-six answers. We are a small design studio in Mumbai, so the honest expectation was a low number.

The number was nine of thirty-six. What matters is the shape of it.

ChatGPT named us in nine of twelve. Perplexity named us in zero of twelve. Gemini named us in zero of twelve.

That is not a twenty-five percent problem. It is a one-engine-in-three problem, and those need completely different work. The two engines on zero describe the studio accurately when asked about it directly. They hold the facts. They just never volunteer them.

Watching what Perplexity actually cited explained why. Every source it used in the blind set was somebody else's roundup article, and we were on none of them. It was not evaluating studios and ranking them. It was reading published lists and repeating the names on them. No amount of markup on our own site reaches that, which is worth knowing before anyone sells structure as the fix for it.

ChatGPT, meanwhile, was doing something closer to evaluating. It read what we publish and formed a view.

Two engines, two entirely different mechanisms, one query. That is the Yext finding, reproduced accidentally at a scale small enough to see the individual answers.

The practical consequence for anyone with a website: "improve our AI visibility" is not a task. It is at least two tasks that share almost nothing.

If the engine reads roundups, your work is off-site. Get named in the lists that actually exist, or publish the measurement those lists get written from. Structured data on your own domain does not touch it.

If the engine evaluates, your work is on-site. Say plainly what you do, who for, and what proof exists. Make it liftable. That work is real and it moves that engine, and it will do nothing for the first one.

Doing only the second and calling it AI visibility is the common mistake, partly because it is the half you control and partly because it is the half most vendors sell.

We published our full run, including the questions where all three engines missed us entirely, at https://reidify.design/ledger. The misses are the useful part. A citation rate with no denominator is a horoscope.

One more thing worth saying plainly. Our thirty-six answers are not a study. They are one studio, one day, one set of questions. What makes them worth publishing is that they point the same direction as seventeen million citations do, and that we recorded them before we knew the result.

My AI visibility score was 78%. Then I fixed how I measured it and it was 25%

reidify — Sat, 22 Aug 2026 10:10:04 +0000

In August I published a number on my studio's site: named unprompted in 14 of 18 blind answers across ChatGPT, Perplexity and Gemini.

It was wrong, and the way it was wrong is worth more than the number.

What I did

I wrote a set of questions and pasted them into one session per engine. Most never mention my studio by name. Two do, because I also wanted to check the engines had the facts right.

You can see it already. Those two questions name the company. Everything in that paste shares one context window. By the time an engine answered question one, the answer was already in front of it.

I was not measuring whether an engine finds my studio unprompted. I was measuring whether it can read.

How I caught it

I did not, at first. I re-ran it, got a near-perfect sweep, and my first reaction was that things had improved.

That reaction is the tell. The output also contained a giveaway: text from one question spliced into the answer to another, mid-sentence. An engine mangling the prompt into the response is not producing a clean read on anything.

The corrected method

Two sessions per engine, not one. Session A holds only the blind questions and never contains the company name. Session B holds the named fact-checks, in a separate fresh incognito session. Three engines, six sessions. Twelve blind questions, so 36 blind answers.

What it actually scored

Nine of 36. Twenty-five percent, against the 78 I had published.

The shape underneath matters more than the total. ChatGPT 9 of 12. Perplexity 0 of 12. Gemini 0 of 12.

It is not a 25% problem. It is a one-engine-in-three problem, and I would never have seen that from the contaminated run, because contamination had lifted all three engines to roughly the same place.

The part I did not expect

Both engines scoring zero describe the studio accurately the moment you ask them directly. They hold the facts. They never volunteer them.

So the problem is distribution, not knowledge and not the website. I checked: every source one of them cited when answering a vendor question was somebody else's roundup. Not one was my own site. It was reading published lists and repeating the names on them, rather than evaluating websites and ranking them.

No amount of schema on my site reaches that.

If you are tracking your own AI visibility

Split the sessions. Never let a question that names you share a context window with a question testing whether you are found.

And if your number goes up after a change, check the method before you celebrate. Mine went up 53 points because I broke the test.

I have published the corrected figure with every miss next to it, along with the twelve questions and the full protocol, at reidify.design/research/measuring-ai-visibility. A flattering number produced by a broken method is worth less than an unflattering one produced by a sound one.

AI Search Is Turning Founder Websites Into Shortlist Evidence

reidify — Mon, 13 Jul 2026 06:37:46 +0000

AI Search Is Turning Founder Websites Into Shortlist Evidence

Search is no longer just a list of blue links. Google is adding AI features and preferred-source behavior. Cloudflare is giving site owners more visibility into AI crawler behavior. ChatGPT is moving more work into one agentic workspace.

For founders, the practical question is not only: can people find the website?

It is: can a person, search assistant, or work agent understand why this business belongs on the short list before the first conversation?

That changes the job of a founder website.

The old website brief was too thin

A normal small-business website often answers four questions:

What do you sell?
Who is it for?
How does it look?
How can someone contact you?

Those still matter. But AI search and source-preference behavior add a fifth question:

Why should this source be trusted enough to carry forward?

That is not solved by adding more generic blog posts. It is solved by making the website more legible as evidence.

What evidence means in practice

A founder website needs to make the business easy to choose, not just easy to browse.

That usually means:

A clear founder point of view.
A visible reason the business exists.
Proof that the team has done real work.
Service pages that explain decision criteria, not just features.
Internal links that connect offer, proof, process, and contact paths.
Copy that says what should happen next.

This is why founder pages and proof pages matter more in the AI-search era. They give humans and machines a cleaner source trail.

The risk for small teams

Small teams often treat the founder page as optional. The homepage gets attention. The service page gets a CTA. The proof page gets added later.

But when AI answers, search summaries, or source-selection tools compress the buying journey, weak trust structure becomes expensive. A vague site may still be crawled, but it may not be selected, cited, summarized, or remembered clearly.

A better website question

Instead of asking whether the website is modern, ask this:

If someone had to decide whether to trust this business from public information alone, what would they quote?

If the answer is unclear, the website needs a stronger evidence layer.

Reidify's founder-led work is built around that layer: clarity, structure, proof, and AI-readable business context. See Rish Sadh at Reidify: https://reidify.design/rish-sadh

The founder website checklist

Before the next redesign, check whether the site answers:

What does the founder believe about the problem?
What work has the business actually shipped?
Which pages prove trust rather than just claim it?
Can an AI system summarize the offer without inventing details?
Is there a clear next step for a serious buyer?

AI search did not make websites less important. It made weak websites easier to skip.

DEV Community: reidify

Search Console now counts searches made with a camera. It cannot tell you what was photographed.

The query column is gone

There is no before

How to read it this week

Every link in an AI Overview gets the same position in Search Console. Google confirmed it.

The position belongs to the box

Impressions miss in both directions

Do not add it to your Web totals

What the report is still good for

How to read it

People Also Ask went from 11% AI to 97% in five months. There is no single answer to own any more.

The numbers, from two sources that did not coordinate

This was not a slow drift

What actually changed about the answer

What to do with that

Google now lets any site leave AI Overviews. Most businesses should not.

What the control actually does

Who it was built for

Why the same trade runs backwards for most businesses

What to do this week

A newsroom just restructured its org chart around losing traffic to AI. Read the memo.

The part that applies past media

What is actually different here

You can now buy ads inside ChatGPT. You still cannot buy a citation.

Google made its own result links unreadable. Rank tracking now costs up to 1,000 requests per query.

The cost is the story

The second problem is worse than the cost

What this does not touch

Google now reports your AI Overviews impressions. Its first drop was a bug.

The first drop in this report was not a drop

Three things that follow

What to do this week

Reddit lost 86% of its ChatGPT citations in four days, and nobody announced it

What this means if you own a website

Four engines, four different answers. We have the receipts for both scales.

My AI visibility score was 78%. Then I fixed how I measured it and it was 25%

What I did

How I caught it

The corrected method

What it actually scored

The part I did not expect

If you are tracking your own AI visibility

AI Search Is Turning Founder Websites Into Shortlist Evidence

AI Search Is Turning Founder Websites Into Shortlist Evidence

The old website brief was too thin

What evidence means in practice

The risk for small teams

A better website question

The founder website checklist