We got tired of guessing.

What gaash.ai actually does

Three things, in order. We measure: a fixed panel of twelve questions your customers would really type, put to Claude with live web search and scored the same way every time. We fix what the measurement found: consistent facts wherever the business is listed, pages that answer the questions in a form a machine can quote, and mentions elsewhere that hold up when someone checks them. Then we measure again once the fixes are live, against the first number rather than anything we said at the outset, and every month after that for clients with monitoring.

What we do not do is as short a list. No secret file to drop on your server. No arrangement with any AI company — none exists to be had. No guaranteed placement in an answer, because the systems that write those answers change weekly and nobody who tells you otherwise has measured it.

Where we came from

Performance marketing, where every claim you make has to survive a number. When customers started asking ChatGPT and Gemini for recommendations instead of typing them into Google, we did what we always do and went looking for the number. Who actually gets named in those answers, and why?

The curiosity turned into a measuring tool, the tool turned into a method, and the method turned into this agency. The unglamorous part is most of the work, and it is the part that moves the number.

The standard: only claim what we can measure

It is an uncomfortable standard and it is the whole company. A claim that cannot be checked is not used here — not in a proposal, not in a report, not on this page. Every engagement starts with a visibility score and is judged against that score, which means that when the number does not move, we say the number did not move.

The same rule governs what we publish. Charts built from example data are labelled as illustrative. Figures carry their source and their date next to the number. When we cite our own research we link the method that produced it, so you can disagree with the method rather than take the figure on trust.

What we have measured so far

Between late August and early September 2026 we ran the same audit against 11,060 small business websites — plumbers, dentists, architects, photographers, car dealerships, forty trades in all, across ninety metros in the US, Canada, Europe and the UAE. 9,483 answered. The point was to find out how much of the AI visibility problem is the assistants and how much is the websites. It is mostly the websites.

One finding changed how we pitch this work. Of every name the assistant gave in those 9,240 probes, 99.4% were actual businesses and 0.6% were directories. Google's first page for the same questions is mostly directories. A small business that has spent ten years failing to outrank an aggregator is being handed a different game, and what decides it is largely what its own site serves.

Source: What 9,483 small-business websites serve to an AI crawler: the method and every figure

What the evidence says works

The academic baseline is a paper called "GEO: Generative Engine Optimization" by Aggarwal and colleagues at Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI, presented at ACM SIGKDD 2024. Tested across live generative engines, the modifications that raised a source's odds of being cited were citing sources, adding statistics and adding quotations — up to around 40% more visibility. Keyword stuffing, the reflex from the last era, did nothing useful.

Our own audit points the same direction from the other end. The sites that got named were the ones with answer-shaped content and an identity a machine could pin down: one name, one address, one phone number, repeated. That is not a trick anyone can sell you. It is a fortnight of unglamorous work, and it is what we do.

Sources: Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (arXiv) · Our audit of 9,483 sites

Who you work with

A small, independent team — no network, no white-label reseller behind us, no media budget to place. The person who runs your audit is the person who answers your email: Hafsa Tariq runs the audits, writes what gaash.ai publishes, and signs the research on this site. Write to hello@gaash.ai and a human replies.

Independence matters more here than it does in most marketing. We have no relationship with OpenAI, Anthropic, Google or Perplexity, no preferential access to any of them, and nothing to sell you that depends on one of them behaving a particular way. What we have is a measurement we run the same way every time and are willing to show you in full.