AI changed keyword research twice at once: models became tools for finding demand patterns humans miss, and AI search became a surface with its own demand, the fan out of sub questions engines generate around every query. Research that covers only the classic keyword list now maps half the territory. From the practice ranked #1 on Google for “Portland Oregon SEO”, checkable from anywhere.
No ranking and no timeline is guaranteed on keyword work, here or anywhere. What is committed to is the turnaround: a prioritised report within 48 hours, naming the highest impact fixes, whether you hire us or not.
Keyword sets are built county by county rather than by radius, and no ranking or timeline is promised for any of them, because nobody controls a search engine.
Keyword sets are scored with the Portland Peak Visibility Score, our own measure of how much of a topic a site actually owns, rather than a raw volume list pulled from a tool.
$199, $950, and $2,500 monthly, month to month, on the pricing page.
Two extensions, both concrete:
The tooling side, used honestly: models are excellent at surfacing question phrasings, intent clusters, and the vocabulary buyers use before they know the term, and unreliable at volumes and facts, so machine suggestions are verified against real data before anything is built. Speed from the model, judgment from the practitioner.
The surface side is the newer work: for each money term, documenting the sub questions answer engines decompose it into, checking which your site already answers at the header level, and assigning the gaps to pages. It is the exact process this site runs on itself, which is checkable in its structure.
Use the model for expansion and classification, and a real data source for volume. That split is the whole discipline. A language model is good at producing the phrasings a customer might use and at sorting them by what the searcher wants; it is not able to tell you how many people searched for any of them last month.
In practice the work runs in five passes:
Most disappointing output comes from a prompt that asked for a list. Ask for groupings and reasoning instead. Give the model the service, the market, the customer, and the constraint, then ask it to group the results by intent and explain each grouping.
“Give me keywords for an electrician” returns the same generic set everyone else gets. “Here is a residential electrician in Washington County who makes most of their money on panel upgrades and emergency callouts. Group the phrases a customer would search by urgency, and say which group is worth building a page for first” returns something usable.
No, and the reason is worth understanding because it is the most expensive mistake in this category. A language model has no access to search volume data. Ask one for monthly search volumes and it will give you numbers. They will look reasonable, they will be formatted convincingly, and they will be invented.
There is a test that takes a minute. Ask the same model for the same volumes twice, in separate sessions. If the numbers move, they were generated rather than retrieved, and nothing in that list should be used to decide where money goes. Confidence is not measurement.
Used correctly the model is doing the part that used to take a week of reading: producing phrasings, spotting that two terms mean the same thing to a buyer, and sorting hundreds of variations into groups a person can act on. That is genuine leverage. It is just not measurement.
By classifying intent at a scale manual review cannot reach. Value is intent multiplied by achievability, not volume. A phrase searched two hundred times a month by people ready to buy is worth more than one searched twenty thousand times by people writing a school assignment, and the second one is usually the one a volume-sorted list puts at the top.
The model is good at telling those apart when you ask it to, and that is the question worth asking it. Keyword sets here are scored on the Portland Peak Visibility Score so the decision rests on how much of a topic a site can realistically own rather than on the largest number in the export.
For patterns and phrasings, useful; for volumes and facts, no, models invent numbers fluently. The service pairs machine breadth with tool verified data, which is the only honest configuration.
The audit includes the check: your money terms fanned out, your headers scanned, and the gaps listed. It is a measurable property of a site, not a vibe, which is why it is in the baseline.
It is included in the monthly tiers rather than sold as a line item: $199 Local, $950 Growth, $2,500 Authority, month to month. The pricing page holds the full breakdown of what each tier weights.
Andreas Benavente, directly. Founder delivery is the product, and it is why the published tiers sit where they do.
The free 48 hour audit: your site crawled, your demand mapped, your baseline recorded, and a prioritised plan written to be usable whether or not you hire anyone.