Principles · Last updated November 3, 2026

How I work, what I won't promise, and what I got wrong

GEO is a young field. Some of what is sold under that name is guesswork with a confident tone. These are the rules I hold myself to, written down so you can check me against them.

The page also lists mistakes I have made on my own site and the rule each one created. I add to it when I find new ones.

Eight working principles

  1. 01

    Measure before I recommend

    Every engagement starts with a baseline: a fixed set of prompts run across ChatGPT, Gemini, Perplexity, Claude and Copilot, plus search data. If I can't measure where you stand today, I can't show you what changed.

  2. 02

    One answer is not a result

    AI answers change between runs, between days and between users. I run each prompt more than once and report how often a brand appears, not a single screenshot that happened to look good.

  3. 03

    No guaranteed citations or rankings

    Nobody controls what an AI engine says. I promise a clear method, honest reporting and work I implement myself. Technical fixes can go live within days; shifts in AI citations usually take 4 to 12 weeks, and sometimes they don't come.

  4. 04

    My own site goes first

    I test methods on yunusmakas.com before I sell them. You can check the result yourself: ask an AI assistant who handles AI visibility in Agder.

  5. 05

    Data, interpretation and forecast stay separate

    In my reports, what was measured, what I think it means and what I expect next are labelled separately. You should always know which sentence is a fact and which is my judgement.

  6. 06

    I implement what I recommend

    Strategy, technical SEO and development are done by one person. There is no handoff where context gets lost, and no recommendation I wouldn't be willing to build.

  7. 07

    Clients decide what gets published

    Client names, numbers and quotes appear on this site only with permission. The Istanbul textile case is anonymous because the client asked for that.

  8. 08

    I say when I don't know

    AI search is new and changes fast. When something is uncertain, untested or outside my experience, I say so instead of filling the gap with confident guesses.

How I use AI in my own work

I use AI tools for development, research and first drafts, including Claude Code for building and auditing sites. They make me faster. They also make confident mistakes.

So every number, claim and code change is checked by me before it reaches a client or this site. An AI tool can suggest a finding; it does not get to approve its own work.

Mistakes I made, and the rule each one created

All four happened on yunusmakas.com, the site I use as proof of my work. Each one is now part of how I audit client sites.

November 2026

Bots read different numbers than people did

The stat counters on my homepage animated from 14 to 20+ years and from 10 to 15 projects. The animation ran only in the browser, so the HTML that crawlers and AI engines read said 14+ and 10. Visitors and machines were told two different stories.

Rule now: I check important pages as raw HTML, the way a bot sees them, not only in a browser.

November 2026

I tracked five engines but named three

I measure ChatGPT, Gemini, Perplexity, Claude and Copilot, but most of my service and process text only mentioned the first three. The two engines with the clearest data sources were the ones missing.

Rule now: What I say I measure is named everywhere I describe the work.

October 2026

My site described my market in two ways

Some pages said I served “Norway and Europe”, others said “Turkey and worldwide”. For an AI system building a picture of who I am, that is a contradiction, and contradictions lower confidence.

Rule now: One canonical description of who I am and where I work, reused in metadata, schema and llms.txt.

October 2026

My sitemap told Bing everything changed, every day

The sitemap set each page’s last-modified date to the build time, so every deploy looked like a site-wide update. Bing learns to ignore that signal on sites where it always changes.

Rule now: Last-modified dates reflect real content changes, and only those pages are submitted through IndexNow.

What I can't promise

These limits apply to every engagement, including mine.

  • Only Microsoft publishes first-party citation data, through the AI Performance report in Bing Webmaster Tools, and it covers Copilot and Bing. For ChatGPT, Gemini, Perplexity and Claude I rely on repeated prompt testing, which is a sample, not a census.
  • AI engines use different indexes. Claude’s web search is powered largely by Brave, Copilot and DuckDuckGo by Bing, and ChatGPT increasingly by its own index. Being visible in one does not mean being visible in all.
  • A citation that exists today can disappear after a model update. That is why ongoing work includes re-measurement, not a one-time report.
  • My Norwegian is functional, not native. Norwegian copy I publish for clients is reviewed by a native speaker.

Hold me to this

If any of my work for you breaks one of these principles, tell me. You can see how I apply them in the case studies, or read more about me.
Start a project

Is your brand visible
in AI search?

Find out how ChatGPT, Gemini, Perplexity, Claude, Copilot and AI search currently see your business, and where the opportunities are.

AI Visibility · Technical SEO · Web Development

Available for new projects

Currently open for AI Visibility, SEO, and web development work.

Phone / WhatsApp

+47 462 46 879

Location

Agder, Norway / Remote

Responsive time

Within 24 hours

Yunus MakasGet in touchStart now