How do I know which SEO and analytics numbers to trust?

By Hema Dey | Founder and CEO, Iffel International Inc. | Forbes Top 5 AI Leader | Author of The AI Translator

Recently I met with a CMO who told me how she measures her website. Clicks. Heat maps. And a comparison against sites that have been doing SEO since before AI search existed. Those are sensible instincts. They were built over a decade in which a click was reliable evidence that someone had considered your business. I respect them, and I want to be direct about why they no longer tell you whether your marketing is working.

The good news is that the new scoreboard is one you can influence. The fundamentals are learnable, and they start with a different question. Not “who clicked?” but “when my buyer asks an AI platform who to trust, does it name us?”

Estimated reading time: 19 minutes

The Click Was a Stand-In. The Answer Is the Product.

For years, the click stood in for consideration. If someone clicked, someone was thinking about you. AI answers break that link, because the thinking now happens on the results page.

SparkToro’s 2026 study found that about 68 percent of Google searches in early 2026 ended without a click to any website (Search Engine Land). Seer Interactive measured what happens when an AI Overview appears: organic click-through rate fell 61 percent, from 1.76 percent to 0.61 percent (Search Engine Land).

Here is the part that matters for revenue. In that same study, brands cited inside the AI Overview earned 35 percent more organic clicks and 91 percent more paid clicks than brands that were not cited. Being cited is not a consolation prize for losing the click. It is how you win the clicks that remain.

Heat Maps Only See the People Who Arrived

Heat maps and session recordings are good tools. They show how visitors behave once they are on a page, and they help you improve conversion for the people you already have.

What they cannot show is the buyer who asked ChatGPT or Google AI Mode who to call, read the answer, and either contacted you or contacted someone else without ever visiting your site. That decision is the one that shapes your revenue, and it happens before your analytics can see it.

The visitors who do arrive from AI platforms are fewer but more valuable. Semrush found that AI search visitors convert at 4.4 times the rate of traditional organic visitors (MarTech). Other studies report multiples from about 4x to more than 20x, depending on method and industry (comparison of the studies), so treat any single number as directional.

A scoreboard built on traffic volume undercounts the channel with higher value per visit. And because AI-influenced visits often arrive without a referrer, they land in your reports as direct traffic, which means even the visits you do get are frequently misattributed.

Why “We Were Doing SEO Before AI” Is Not a Benchmark

The second comparison she made was between sites that have been doing SEO for years and newer sites built in 2026. It is a natural comparison, and it is the wrong test.

Age and link history are stand-ins for something else: accumulated validation. NP Digital’s analysis of ChatGPT citations found that older sites appear more often, largely because they have collected brand mentions and authority over time, while newer sites are underrepresented but not excluded (NP Digital). Ahrefs research found that branded web mentions had the strongest relationship with AI brand visibility, well ahead of backlinks (summarized by Loganix).

So the question is not how old the site is. It is what an AI can verify about the business behind it. A 2026 site with a consistent entity record, named expert authors, reviews on independent platforms, and independent press can be recommended. A ten-year-old site with thin validation can be passed over, however many years of SEO sit behind it.

The older site does have a head start, and head starts compound. That is a reason to start building validation now, not a reason to wait. The fair benchmark is your competitors, on the specific questions your buyers ask, measured by whether AI names you.

What AI Search Actually Runs On: Validation Signals

AI platforms do not rank pages the way a results page does. They assemble an answer and decide which businesses they can name with confidence. If an AI cannot confirm who you are, what you are known for, and who else says so, the safe move is to leave you out.

This is already visible in the data. A study reported by Search Engine Journal found that AI systems recognized 96 percent of the brands measured, yet 89 percent never appeared in AI answers to category research questions (Search Engine Journal). Recognition is not recommendation. Validation is the difference.

Where does validation come from? AirOps research found that about 85 percent of brand mentions in AI responses come from third-party pages rather than the brand’s own site (reported by Superlines). Methods differ across studies. Yext, for example, counts listings and reviews as brand-managed sources and finds that most citations come from those (Chain Store Age). Every study still points the same way: AI leans on evidence that your own homepage cannot supply alone.

At Iffel International we organize that evidence around the 4 Vs from The AI Translator:

• Visibility: Can machines read you? Structured data, crawlable pages, and consistent technical health.
• Validity: Is the same true story told everywhere AI looks?
• Veracity: Do named experts and independent sources confirm your expertise?
• Values: Does your reputation, including your reviews, exist as evidence beyond your own voice?

When the signals in any one of these are weak or missing, AI hesitates. When several are weak, the answer to “why are we not being recommended?” is simple. You are invisible.

The People Behind the Brand: Trust Is Personal, Especially for YMYL

When a topic touches health, money, or legal outcomes, Google calls it YMYL, short for Your Money or Your Life, and holds it to a higher standard. Google’s Search Quality Rater Guidelines place trust at the center of E-E-A-T: a page that lacks trust has low E-E-A-T however experienced or expert it appears. The guidelines also tell raters to research the reputation of the company and of the content creator, including looking across other websites to find it.

One honest caveat. The guidelines are instructions for human raters, not a ranking formula. They describe what Google’s systems are meant to reward, and that is reason enough to take them seriously.

For a law firm, a medical practice, or a financial advisory, the person is the brand. The attorney, physician, or advisor whose name is on the page is the evidence the page rests on. And buyers now check that evidence through AI. BrightLocal’s 2026 survey found that 45 percent of consumers now use ChatGPT or other generative AI tools for local business recommendations, up from 6 percent a year earlier. A vendor-run study of 10,000 local businesses found that those recommended by ChatGPT and Perplexity averaged 133 Google reviews, while those AI ignored averaged 11. That is a correlation, not proof of cause, but it points the same way as everything else in this post.

So here is the question I would ask every leadership team. What do the reviews and profiles of each person we put forward say about us? I have not found a study that isolates how one person’s reviews affect a firm’s AI recommendations, so treat this as reasoned inference rather than a measured effect. But when a brand is presented through named people, each person’s reputation becomes part of the evidence that raters and AI systems read about the brand. A weak pattern attached to one or two names can pull down the sentiment attached to the whole.

Here is how I would audit it:

1. List everyone your site names. Partners, physicians, advisors, and authors.
2. Check where a buyer or an AI would look. Google reviews that mention each person by name, the industry profiles for your field (for example Avvo, Martindale, Healthgrades, or Zocdoc), LinkedIn, news mentions, and the licensing board record where relevant.
3. Look for patterns, not one-offs. Repeated complaint themes, unanswered reviews, a long gap since the last review, or a profile that contradicts the bio on your site.
4. Fix the cause before the display. If reviews point to a service problem, address it through coaching or a process change. Do not hide it, and do not feature someone more prominently until the pattern changes.
5. Respond to reviews promptly and professionally, without confirming that someone is a client or patient, so you protect confidentiality.
6. Keep each person’s bio, credentials, and profile links consistent with the Person schema on your site.

One related caution. Do not mark up your own reviews on your own site to try to earn star ratings in Google. Google treats them as self-serving and no longer displays review snippets for LocalBusiness and Organization pages. The work is earning real reviews on independent platforms, not decorating your own.

What to Avoid When You Launch a New Website

A new site has no history to lean on, which makes every signal it sends count for more. In our audits, much of the damage on new sites is not a lack of effort. It is confusion: signals that disagree with each other, so an AI system has no single clear picture of the business and takes the safe route of naming someone else.

Avoid	Why it creates confusion	Do instead
A domain that does not match your business name	Your LinkedIn, Google Business Profile, press, and reviews use one name and your URL says another, so a machine has to guess they are the same business. Google’s site name system weighs your structured data, title tag, headings, and og:site_name, and falls back to showing the domain when it cannot identify a clear name.	Use your business name, or an obvious variant, as the domain. Use that exact name in WebSite schema, the title tag, headings, and og:site_name. If you change domains, 301 redirect the old one and update every profile.
Several versions of the same site	http and https, www and non-www, an old domain and a new one, or a staging copy left live split your signals across copies. A copy you never meant to publish can be the one that gets chosen.	Pick one preferred version and 301 redirect the rest. Keep staging password-protected. Make sure the WordPress setting that discourages search engines is off on the live site and on for staging.
Canonical tags that contradict the page	A canonical pointing at a staging domain, another page, or the homepage tells search engines the page is a duplicate of something else. Google weighs a canonical against redirects, internal links, and sitemaps, and can choose a different URL than the one you declared.	Give every indexable page a self-referencing canonical that matches its URL in your sitemap and internal links. Audit after every launch and migration.
Descriptions that disagree with each other	Your meta description, social description, schema description, About page, LinkedIn, and Google Business Profile each describe a different service or position. An AI summarizing you has no single version to trust.	Write one approved description in three lengths, short, medium, and full, and reuse it everywhere.
A name, address, phone number, or title that varies	To a machine, Acme Legal, Acme Legal Group LLP, and Acme are three different names. Old addresses and outdated founder titles add more.	Keep one source-of-truth sheet and correct your top listings first.
Schema that contradicts the visible page	Duplicate Organization entities from stacked plugins, the wrong business type, or a name in schema that differs from the name on the page. This is a recurring finding in our audits.	One Organization entity, one name. Validate with the Rich Results Test and the Schema Markup Validator after every update.
Blocking the crawlers you want to read you	A firewall bot wall, a robots.txt left over from staging, or an aggressive caching rule can hide a new site from the crawlers it depends on.	Test what Googlebot and AI crawlers can actually fetch before launch and after every security or caching change.
Launching with no human behind it	No named authors, no team or About page, and no verifiable credentials leave AI unable to tell who is accountable for your claims.	Name your experts, link to their profiles, and add Person schema so authorship can be verified.

A note on canonicals, because they are a quiet way to lose. Google’s documentation treats a canonical tag as a hint to be weighed against other signals, and it tells site owners not to declare one URL in the sitemap and a different one in the canonical tag (Google Search Central). In a February 2026 case study, SEO consultant Glenn Gabe found that when Google ignored a site’s canonical, ChatGPT surfaced the URL Google had picked (reported by GEO Toolbox). If your signals disagree, the page a buyer is sent to may not be the one you built.

None of this is dramatic. It is hygiene, and it is far cheaper to get right at launch than to repair after AI platforms have already formed a view of you.

The Scoreboard to Use Instead

Old scoreboard	What it measures	New scoreboard	What it tells you
Clicks and sessions	Visits from people who chose to click	AI citation share	How often AI platforms name your business in answers to your buyers’ questions
Keyword rankings	Position on a results page	Mentioned, cited, recommended	Whether AI names you, links to you, or tells the buyer to contact you. Three outcomes, three different revenue values
Heat maps and scroll depth	Behavior of people who already arrived	Heat maps for conversion work only	Keep using them to improve pages. Stop using them as proof of visibility
Backlink counts	Links accumulated over time	Independent mentions (frontlinks)	Whether credible third parties confirm your expertise
Pageviews	Volume	AI-referred leads and revenue	Whether AI-influenced buyers become clients

What Data to Trust, and What to Treat as Directional

That CMO’s scoreboard came from dashboards, so it is worth being clear about which numbers in them are measurements and which are estimates. First-party tools record what happened on your own property. Third-party tools model what they cannot see. Both are useful, and neither is a census.

Review scores below were checked in November 2026, using the latest figures each platform displayed. They move over time.

Tool	Reviews	Trust it for	Treat with caution
Google Search Console First-party	G2 4.7 (about 510 reviews). Capterra about 4.8 (about 218). Reviewers value a free direct line to Google. Complaints: report complexity and data limits.	Your actual Google clicks and impressions, indexing status, and which URL Google chose as canonical. A direct record of what Google sent you.	In an Ahrefs analysis, about 47 percent of clicks came from anonymized queries, so query-level data is incomplete. The interface shows 1,000 rows per table. AI Mode activity has been folded into Web totals with no filter, and the June 2026 generative AI report shows impressions only, without clicks or queries.
Google Analytics 4 First-party	G2 4.5 (about 6,860 reviews). Reviewers value it as a free standard. The common complaint is GA4’s steep learning curve.	What visitors do after they arrive: conversions, engagement, and trends over time.	It is not a full count. Large reports can be sampled or thresholded, and ad blockers and declined consent mean some visits are modeled or missing. AI-influenced visits often arrive without a referrer and are counted as direct.
Microsoft Clarity First-party, heat maps	G2 4.5 (55 reviews). Capterra 4.8 (56). A small review base, with praise for free heat maps and recordings.	Seeing where visitors hesitate, rage-click, or drop off on pages they reached.	Like every on-site tool, it only sees people who arrived and loaded the script. It cannot see the buyer who never visited.
Semrush Third-party estimates	G2 4.4 (about 4,000 reviews). Capterra 4.6 (about 2,330). A far larger review base than most tools here. Complaints: price and learning curve.	Keyword research, competitor comparisons, rank tracking, and relative trends.	Traffic figures are modeled estimates. Collaborator.pro’s 184-site test measured an average error of about 62 percent, with a tendency to overestimate. Screaming Frog’s 25-site UK test found total traffic underestimated by about 30 percent.
Ahrefs Third-party estimates	G2 4.5 (about 710 reviews). Capterra 4.7 (585). Reviewers often praise its data quality. Complaint: price.	Backlink and mention research, competitor analysis, and tracking brand mentions across AI platforms with Brand Radar, added in 2026.	Its estimates are directional too. In Collaborator.pro’s test it had a lower average error than Semrush or Similarweb, about 49 percent, and tended to underestimate. Ahrefs’ own 1,635-site study put its median error near 50 percent. Reviewers report estimates wobble for very small sites.
Ubersuggest Third-party estimates	G2 4.2 (about 160 reviews). Capterra 4.4 (93). A much smaller review base than the paid suites, with praise for price and ease of use.	Keyword ideas, quick competitor checks, and prompt-level AI visibility tracking for small teams and beginners.	I found no rigorous independent accuracy study of its traffic estimates. Roundups describe them as directional. Use it for ideas and trends, not for numbers you report as fact.

Two cautions when you read that table. First, review scores measure satisfaction, not accuracy. A tool can be well liked and still be off by half on traffic. Second, the widely cited accuracy comparisons were run by vendors or agencies with a stake in the answer, so treat them as indicative. The consistent finding is that third-party traffic numbers can miss real traffic by a wide margin, in either direction.

Here is how I use them together:

• Search Console and GA4 answer different questions. Search Console is the record of what Google sent you. GA4 is the record of what visitors did next. When they disagree, expect it, because they count different things.
• Use Semrush, Ahrefs, and Ubersuggest for direction, not for facts. Compare yourself to competitors and watch the trend. Never quote a third-party traffic figure as if it were real, and show a range when you must.
• Use heat maps to improve pages after people arrive. Do not use them to measure whether you are visible.
• Treat every AI visibility tracker as a trend line. AI answers change from run to run. In AirOps data reported by AuthorityTech, only about 30 percent of brands stayed present across consecutive AI answers, and fewer than 20 percent persisted across five runs. Run each prompt several times and watch the direction over weeks, not a single result.
• Add the one source no tool can supply. Ask new clients how they found you. That answer is first-party data about the decision that happens before anyone clicks.

A Weekly Habit That Beats Any Report: Read Your Leads

Reports look backward, and as the data section shows, they miss a growing share of AI-influenced visits. The clearest read on your pipeline is the leads themselves. So here is a hygiene habit I would put ahead of any new dashboard: once a week, the people responsible for growth read the actual leads together.

Not a summary. The leads. Thirty minutes, the same day each week, with marketing, sales or intake, and a leader who owns revenue in the room.

1. Read every new lead from the week. For each one, note where they say they found you (including AI assistants), what they asked for, whether it was a fit, and what happened next.
2. Sort real leads from time-wasters. Define “real” in one sentence for your business first: the right service, the right location, able to buy, reachable. Then ask why each time-waster got through. A vague page, a loose form, or a keyword that attracts the wrong person usually explains it.
3. Fix one thing per week. A page, a form question, an FAQ. Small weekly corrections compound faster than a quarterly overhaul.
4. Celebrate wins openly. Name the lead that became a client and what brought them in. Share it with the whole team. Wins show people what good looks like, and they tell you which signals are working.
5. Keep a simple tally. New leads, real leads, booked, won, and top sources including AI. Watch the trend over weeks, not days.

Here is why this matters now. Because AI-referred visitors tend to arrive more informed, the first sign of progress may be lead quality rather than lead volume. A report built on volume will miss it. A person reading the leads will not. The shift from time-wasting leads to real leads is one of the earliest signs that your validation signals are working.

The businesses that make this a weekly rhythm build something no tool provides: a culture that checks the pulse of the business together, talks about it openly, and celebrates what is working. That culture is what sets the front runners apart. In legal and healthcare settings, keep the review inside your confidentiality rules and remove identifying details when you share a win.

What I Would Do on Monday

1. Build a baseline. Write down the ten questions your best clients ask before they call. Run them in ChatGPT, Gemini, Google AI Mode, and Perplexity. Record whether you and two competitors are mentioned, cited, or recommended.
2. Change how you ask about attribution. Add “how did you first hear about us?” to every intake call and form, with AI platforms as an option. Track sales conversations and lead quality, not only sessions.
3. Start the weekly lead review. Put thirty minutes on the calendar, read the actual leads together, and celebrate what is working.
4. Audit your validation signals. Check your schema, your entity consistency across every listing and bio, your named authors, and whether your reviews on independent platforms are current and consistent.
5. Audit the people behind the brand. Review the reputation of every named expert, and fix the cause of any weak pattern before you feature them more prominently.
6. Build independent evidence. Earn press, speaking credits, directory presence, and partner mentions that confirm what your own site claims.
7. Keep the fundamentals and move the scoreboard. Keep your SEO foundations and your heat maps for what they do well. Report monthly on AI citation share, AI-referred leads, and revenue influenced.

The Risk Is Gradual, Which Is Why It Gets Missed

Steering by clicks does not fail loudly. Rankings can hold, impressions can climb, and the report can look healthy while fewer people click and more decisions are made upstream, in answers your report cannot see. Meanwhile a competitor with stronger validation signals is named again and again, and that presence compounds.

This is not a case against SEO. Crawlable, fast, well-structured pages are the foundation AI platforms read from, and strong SEO work supports strong GEO. It is a case against treating clicks as the finish line. A business that returns to the old scoreboard and calls it strategy is optimizing for a shrinking share of how buyers decide. Left long enough, that is a risk to revenue and, eventually, to survival. It is also entirely avoidable.

Clicks were never the goal. Being chosen was. The AI era simply makes that visible.

If you want to see where your business stands, I will run the baseline with you: your buyers’ questions, the major AI platforms, the three outcomes, and a clear picture of who AI is naming instead of you. Start at iffelinternational.com/contact-us. To go deeper on the framework, see GEO2Sales, SEO2Sales, and Why AI Is Not Recommending Your Brand.